Coaxial Electrical
Circuits for
InterferenceFree
Measurements
Other volumes in this series:
Volume 4 The current comparator W.J.M. Moore and P.N. Miljanic
Volume 5 Principles of microwave measurements G.H. Bryant
Volume 7 Radio frequency and microwave power measurement A.E. Fantom
Volume 8 A handbook for EMC testing and measurement D. Morgan
Volume 9 Microwave circuit theory and foundations of microwave metrology
G. Engen
Volume 11 Digital and analogue instrumentation: testing and measurement
N. Kularatna
Volume 12 Microwave measurements, 3rd edition R. Collier and D. Skinner (Editors)
Coaxial Electrical
Circuits for
InterferenceFree
Measurements
Shakil Awan, Bryan Kibble
and Ju rgen Schurr
The Institution of Engineering and Technology
Published by The Institution of Engineering and Technology, London, United Kingdom
The Institution of Engineering and Technology is registered as a Charity in England & Wales
(no. 211014) and Scotland (no. SC038698).
2011 The Institution of Engineering and Technology
First published 2011
This publication is copyright under the Berne Convention and the Universal Copyright
Convention. All rights reserved. Apart from any fair dealing for the purposes of research
or private study, or criticism or review, as permitted under the Copyright, Designs and
Patents Act 1988, this publication may be reproduced, stored or transmitted, in any
form or by any means, only with the prior permission in writing of the publishers, or in
the case of reprographic reproduction in accordance with the terms of licences issued
by the Copyright Licensing Agency. Enquiries concerning reproduction outside those
terms should be sent to the publisher at the undermentioned address:
The Institution of Engineering and Technology
Michael Faraday House
Six Hills Way, Stevenage
Herts, SG1 2AY, United Kingdom
www.theiet.org
While the authors and publisher believe that the information and guidance given in
this work are correct, all parties must rely upon their own skill and judgement when
making use of them. Neither the authors nor publisher assumes any liability to
anyone for any loss or damage caused by any error or omission in the work, whether
such an error or omission is the result of negligence or any other cause. Any and all
such liability is disclaimed.
The moral rights of the authors to be identified as authors of this work have been
asserted by them in accordance with the Copyright, Designs and Patents Act 1988.
British Library Cataloguing in Publication Data
A catalogue record for this product is available from the British Library
ISBN 9781849190695 (paperback)
ISBN 9781849190701 (PDF)
Typeset in India by MPS Ltd, a Macmillan Company
Printed in the UK by CPI Antony Rowe, Chippenham, Wiltshire
Contents
List of figures xiii
Abstract xxiii
Preface xxv
Acknowledgements xxvii
1 Introduction 1
1.1 Interactions between circuits eliminating electrical interference 1
1.1.1 Basic principles 2
1.1.2 An illustrative example using a phasesensitive detector 11
1.1.3 Diagnostic equipment 12
1.1.4 Isolation 17
1.1.5 Totally isolating transformers and power supplies 18
1.1.6 Isolating a noisy instrument 20
1.1.7 The available methods for isolating outputs 22
1.1.8 Balancing 24
1.1.9 Minimising the effects of insufficiently isolated
commercial instruments 24
1.1.10 The traditional approach to DC and lowfrequency
circuitry versus the currentbalanced conductorpair
coaxial approach 26
1.1.11 Thermoelectric emfs 27
1.1.12 Designing temperaturecontrolled enclosures 28
1.1.13 Ionising radiation (cosmic rays, etc.) 30
1.1.14 Final remarks 31
References 31
2 Sources, detectors, cables and connectors 33
2.1 General principles 33
2.1.1 The output impedance of a network affects
detector sensitivity 33
2.1.2 The sensitivity of detectors to harmonic content 34
2.1.3 Noise and noise matching a detector to a network 37
2.1.4 The concept of a noise figure 39
2.2 Attributes of sources 41
2.3 Properties of different detectors 42
2.3.1 Preamplifiers 42
2.3.2 Wideband (untuned) detectors 43
2.3.3 Narrowband (tuned) detectors 44
2.3.4 Phasesensitive detectors that employ a
switching technique 44
2.3.5 Phasesensitive detectors employing a
modulating technique 44
2.4 Cables and connectors 45
References 46
3 The concept of a lowfrequency coaxial network 47
3.1 The coaxial conductor 49
3.1.1 Achieving current equalisation 49
3.1.2 The concept of a coaxial network 53
3.2 Construction and properties of coaxial networks 54
3.2.1 Equalisers in bridge or other measuring networks 54
3.2.2 Assessing the efficiency of current equalisers 57
3.2.3 Single conductors added to an equalised network 59
3.2.4 Other conductor systems having similar properties 61
3.2.5 DC networks 63
3.2.6 The effect of a length of cable on a measured value 63
3.2.7 Triaxial cable 67
References 68
4 Impedance measurement 69
4.1 Improvements in defining what is to be observed or measured 69
4.1.1 Ratio devices 70
4.1.2 Impedance standards 70
4.1.3 Formal representation of circuit diagrams and components 73
5 General principles of accurate impedance measurement 79
5.1 The evolution of a coaxial bridge 79
5.1.1 A simple coaxial bridge as an example of a coaxial
network 82
5.2 The validity of lumped component representations 84
5.3 General principles applying to all impedance standards 85
5.3.1 The physical definition of a standard 85
5.3.2 The electrical definition of a standard impedance 86
5.3.3 Twoterminal definition 87
5.3.4 Fourterminal definition 88
5.3.5 Fourterminal coaxial definition 88
5.3.6 Twoterminalpair definition 88
5.3.7 Threeterminal definition 89
5.3.8 Fourterminalpair definition 90
vi Coaxial electrical circuits for interferencefree measurements
5.3.9 Measuring fourterminalpair admittances in a
twoterminalpair bridge by extrapolation 92
5.3.10 Adaptors to convert a two or fourterminal
definition to a fourterminalpair definition 94
5.4 The effect of cables connected to the ports of impedance standards 95
5.4.1 The effect of cables on a twoterminal component 96
5.4.2 The effect of cables on a fourterminal coaxial component 96
5.4.3 The effect of cables on a twoterminalpair component 99
5.4.4 The effect of cables on a fourterminalpair component 100
5.5 An analysis of conductorpair bridges to show how the effect
of shunt admittances can be eliminated 102
5.5.1 Comparing direct admittances using voltage sources 103
5.6 Combining networks to eliminate the effect of unwanted
potential differences 104
5.6.1 The concept of a combining network 104
5.6.2 A general purpose AC combining network and
current source 105
5.7 Connecting twoterminalpair impedances in parallel 107
References 107
6 Impedance standards 109
6.1 The history of impedance standards 110
6.2 The ThompsonLampard theorem 111
6.3 Primary standards of phase angle 114
6.4 Impedance components in general 115
6.4.1 Capacitors 115
6.4.2 Parallelplate capacitance standard 117
6.4.3 Twoterminal capacitors 122
6.4.4 Threeterminal capacitors 122
6.4.5 Two and fourterminalpair capacitors 123
6.4.6 The mechanical construction and properties of various
types of capacitors 123
6.4.7 Capacitance standards of greater than 1 mF 126
6.4.8 Voltage dependence of capacitors 129
6.4.9 Resistors 130
6.4.10 Tnetworks 132
6.4.11 Adding auxiliary components to resistors to reduce
their reactive component 133
6.4.12 Mutual inductors: Campbells calculable mutual
inductance standard 135
6.4.13 Selfinductors 139
6.5 Resistors, capacitors and inductors of calculable frequency
dependence 140
6.5.1 Resistance standards 140
6.5.2 Haddad coaxial resistance standard 140
Contents vii
6.5.3 A nearly ideal HF calculable coaxial resistance standard 144
6.5.4 A bifilar resistance standard 146
6.5.5 Gibbings quadrifilar resistance standard 147
6.5.6 Bohacek and Wood octofilar resistance standard 149
6.5.7 HF secondary resistance standards 150
6.5.8 HF parallelplate capacitance standard 152
6.5.9 HF calculable coaxial capacitance standard 153
6.5.10 HF calculable coaxial inductance standard 155
6.5.11 A frequencyindependent standard of impedance 156
6.5.12 An ideal standard of impedance of calculable frequency
dependence 157
6.6 Quantum Hall resistance 162
6.6.1 Properties of the quantum Hall effect (QHE) and its
use as a DC resistance standard 162
6.6.2 The properties and the equivalent circuit
of a quantum Hall device 166
6.6.3 Device handling 171
6.7 QHE measured with AC 171
6.7.1 Multipleseries connection scheme 173
6.7.2 A device holder and coaxial leads 176
6.7.3 Active equalisers 179
6.7.4 Capacitive model of ungated and splitgated
quantum Hall devices 180
6.7.5 Ungated quantum Hall devices 181
6.7.6 Splitgated quantum Hall devices 182
6.7.7 Doubleshielded device 183
References 184
7 Transformers 187
7.1 General considerations 188
7.1.1 The causes of departure from an ideal transformer 188
7.1.2 The magnetic core 189
7.1.3 The windings; the effect of leakage inductances,
capacitances and resistances 192
7.1.4 Representation of a nonideal transformer: the effect
of loading on its ratio windings 195
7.1.5 The twostage principle 197
7.1.6 Electrical screens between windings 199
7.2 Constructional techniques 200
7.2.1 Design of transformer windings 200
7.2.2 Techniques for minimising the effect of leakage
inductance, winding resistance and the capacitances
of ratio windings 203
7.2.3 Bifilar winding 204
7.2.4 Rope winding having randomly arranged strands 205
viii Coaxial electrical circuits for interferencefree measurements
7.2.5 Ordered rope winding 206
7.2.6 Magnetic and electric screens 207
7.2.7 Testing the attainment of a nearly toroidal field 208
7.2.8 Connections to the output ports 209
7.3 Types of transformers 210
7.3.1 Inductive voltage dividers 210
7.3.2 Twostaged IVDs 216
7.3.3 Injection and detection transformers 218
7.3.4 Use of an injection transformer as a small voltage source 220
7.3.5 Use of an injection transformer as a detector of zero current 222
7.3.6 Calibration of injection transformers and their associated
phase change circuits 222
7.3.7 Voltage ratio transformers 224
7.3.8 Twostage construction 225
7.3.9 Matching transformers 226
7.3.10 Current ratio transformers 226
7.3.11 Highfrequency construction 227
7.4 Calibration of transformers 228
7.4.1 Calibrating an IVD in terms of a fixedratio transformer 228
7.4.2 Calibrating voltage ratio transformers using a calibration
transformer with a single output voltage 229
7.4.3 Calibration with a 1:1 ratio transformer 232
7.4.4 The bridge circuit and details of the shielding 236
7.4.5 The balancing procedure 238
7.4.6 Calibrating voltage transformers by permuting
capacitors in a bridge 240
7.4.7 Calibration of current transformers 242
7.4.8 Assessing the effectiveness of current equalisers 243
References 243
8 General considerations about impedance comparison networks 245
8.1 Designing bridge networks 245
8.1.1 Applying coaxial techniques to classical
singleconductor bridges 245
8.1.2 Placement of current equalisers 246
8.1.3 Wagner circuit (and when it is applicable) 246
8.1.4 Convergence 247
8.1.5 Moving a detector to other ports in a bridge network 249
8.1.6 Tconnecting shunt impedances for balance
adjustment 249
8.1.7 Role of electronics in bridge design 250
8.1.8 Automating bridge networks 250
8.1.9 Higherfrequency networks 250
8.1.10 Tests of the accuracy of bridges 251
References 253
Contents ix
9 Bridges to measure impedance ratios 255
9.1 Bridges to measure the ratio of like impedances 256
9.1.1 A twoterminal IVD bridge 256
9.1.2 A twoterminalpair IVD bridge 257
9.1.3 A fourterminalpair IVD bridge 258
9.1.4 A twoterminalpair bridge based on a 10:1 voltage
ratio transformer 260
9.1.5 A fourterminalpair bridge based on a twostage
10:1 voltage ratio transformer 261
9.1.6 Equalpower resistance bridge 263
9.2 Bridges to measure the ratio of unlike impedances 263
9.2.1 RC: the quadrature bridge 263
9.2.2 The quadrature bridge a twoterminalpair design 267
9.2.3 The quadrature bridge a fourterminalpair design 268
9.2.4 Bridges for measuring inductance 271
9.3 AC measurement of quantum Hall resistance 272
9.3.1 AC contact resistance 272
9.3.2 AC longitudinal resistance 273
9.3.3 Measuring R
xxLo
274
9.3.4 Measuring R
xxHi
276
9.3.5 A simple coaxial bridge for measuring nondecade
capacitances 280
9.3.6 Coaxial resistance ratio bridges involving quantum
Hall devices 281
9.3.7 A quadrature bridge with two quantum Hall devices 282
9.4 Highfrequency networks 283
9.4.1 An IVDbased bridge for comparing 10:1 ratios of
impedance from 10 kHz to 1 MHz 283
9.4.2 A bridge for measuring impedance from 10 kHz to
1 MHz based on a 10:1 voltage ratio transformer 286
9.4.3 Quasifourterminalpair 1:1 and 10:1 ratio bridges for
comparing similar impedances from 0.5 to 10 MHz 286
9.4.4 A fourterminalpair 10MHz 1:1 resistance ratio bridge 288
9.4.5 A 1.6 and 16MHz quadrature bridge 289
9.4.6 Fourterminalpair resonance frequency measurement
of capacitors 291
9.4.7 Scattering parameter measurements and the link to
microwave measurements 292
9.4.8 Electronic fourterminalpair impedancemeasuring
instruments 295
References 296
10 Application of interferencefree circuitry to other measurements 299
10.1 Resistance thermometry (DC and lowfrequency AC) 299
10.1.1 DC resistance thermometry 299
10.1.2 AC resistance thermometry 299
x Coaxial electrical circuits for interferencefree measurements
10.2 Superconducting cryogenic current comparator 300
10.2.1 Determining the DC ratio of two resistances R
1
/R
2
300
10.3 Josephson voltage sources and accurate voltage measurement 302
10.4 Future directions 304
10.4.1 Higherfrequency measurements of quantum Hall resistance 304
10.4.2 Comparing calculable resistance standards up to
100 MHz with finiteelement models 304
10.4.3 Radiofrequency and microwave measurements of carbon
nanotubes and graphene 305
References 306
Appendix 1 307
Appendix 2 308
Appendix 3 309
Appendix 4 310
Index 313
Contents xi
Figures
Figure 1.1 Coupling via two capacitances 3
Figure 1.2 Coupling via one capacitance and a conductor 3
Figure 1.3 Placing a screen to eliminate capacitive coupling 3
Figure 1.4 (a) A conducting control shaft penetrating a shield.
(b) Connecting the shaft to the shield with a spring contact 4
Figure 1.5 Capacitances between conductors on a printed
circuit board 4
Figure 1.6 A ground plane on the reverse side of a printed circuit board 5
Figure 1.7 Inductive (magnetic) coupling 5
Figure 1.8 Eliminating inductive coupling 5
Figure 1.9 Amperes theorem applied to a circular path concentric
with the axis of a coaxial cable 6
Figure 1.10 (a) A toroidal magnetic shield surrounding a toroidal
magnetic core. (b) A practical construction for the shield 6
Figure 1.11 Commonconductor coupling 7
Figure 1.12 A star connection to eliminate commonconductor coupling 7
Figure 1.13 A branch connection to eliminate commonconductor coupling 7
Figure 1.14 The currents flowing in an unsymmetrical amplifier
circuit 8
Figure 1.15 The currents in a twostage operational amplifier circuit.
Heavy lines indicate occurrences of commonconductor
coupling 9
Figure 1.16 Removing commonconductor couplings from the
circuit of Figure 1.15 9
Figure 1.17 S is the vector sum of several (A, B, C, etc.)
phaserelated interferences from a common source.
S
0
is the new vector sum if C is eliminated 10
Figure 1.18 Connecting a phasesensitive detector to a network 11
Figure 1.19 An isolating transformer can break the loop 12
Figure 1.20 An added conductor and an equaliser reduce the
current in the outer conductor of the input cable
to the detector 13
Figure 1.21 A differentialinput preamplifier 14
Figure 1.22 A pickup coil for detecting magnetically coupled interference 15
Figure 1.23 Probes to detect or produce electric fields 15
Figure 1.24 An isolating enclosure with a filtered output 18
Figure 1.25 A totally isolated mainsdriven power supply 19
Figure 1.26 A cross section through a toroidal transformer
for a totally isolated supply 19
Figure 1.27 The best way to overlap nested toroidal shields 20
Figure 1.28 A typical commercial instrument as a noise source 21
Figure 1.29 Isolating a noisy instrument 21
Figure 1.30 An isolation amplifier where one direct connection is
permissible 23
Figure 1.31 Balancing a conductorpair circuit 25
Figure 1.32 An example of a small temperaturecontrolled
enclosure 29
Figure 1.33 Two thermistors in a Wheatstone bridge for
temperature sensing 30
Figure 2.1 The output impedance of a network and the shunt
admittance of an input cable and detector affect
sensitivity 34
Figure 2.2 A twinT bridge 34
Figure 2.3 (a) A harmonicrejecting filter. All cores are highQ
amorphous metal material. (b) Its frequency response 36
Figure 2.4 An ultralow noise harmonicrejecting filter for use
with a bridge having components at cryogenic temperatures 37
Figure 2.5 (a) A component coated with a lossy dielectric.
(b) Removing the loss contribution to the noise 38
Figure 2.6 The equivalent circuit of a detector connected to
a noise source 38
Figure 2.7 Circuits to match a detector to a network for better sensitivity:
(a) a source, the equivalent circuit of a network, the equivalent
noise representation of a preamplifier and a final detector;
(b) and (c) matching a lowimpedance source and network
to a highimpedance detector; (d) matching a highimpedance
source and network to a lowimpedance detector;
(e) an isolating matching transformer 40
Figure 2.8 The equivalent circuit of a detector 41
Figure 2.9 A lownoise preamplifier suitable for highimpedance
sources 42
Figure 2.10 A lownoise preamplifier suitable for source impedances
of the order of 10 kW 43
Figure 3.1 (a) One mesh of a coaxial network threaded through
a core and (b) the equivalent circuit 50
Figure 3.2 (a) A highpermeability toroidal transformer core used as a
current equaliser. (b) Its symbolic representation 53
Figure 3.3 An equalisedcurrent conductorpair mesh 53
Figure 3.4 A coaxial version of Figure 3.3 54
Figure 3.5 An active current equaliser 56
Figure 3.6 Circuit details of an active current equaliser 56
xiv Coaxial electrical circuits for interferencefree measurements
Figure 3.7 Evaluating the effectiveness of a current equaliser:
(a) the injection and detection circuits; (b) a practical
inphase and quadrature quasicurrent source 58
Figure 3.8 (a) The network and (b) the nodes of the circuit of outer
conductors. (c) In the network of cable without equalisers 60
Figure 3.9 A single conductor added to a currentequalised network 61
Figure 3.10 The magnetic field surrounding a twistedpair cable 62
Figure 3.11 Interconnection of the shields of twistedpair cables 62
Figure 3.12 The equivalent circuit of a coaxial cable 64
Figure 3.13 The relationship between the currents at the ends of a cable 65
Figure 3.14 The relationship between the voltages at the ends of
an opencircuited cable 65
Figure 3.15 The combined effect of two cables in series 66
Figure 3.16 A triaxial cable with the intermediate shield raised
to the same potential as the inner conductor 67
Figure 4.1 The direct admittance or impedance of a component 71
Figure 4.2 Shunting a fourterminal resistor via extra connections
to its internal defining points 72
Figure 4.3 Shunting a fourterminal resistor (R
Sh
)R) 72
Figure 4.4 A direct very high impedance constructed from a
Tnetwork of lowervalue impedances 73
Figure 4.5 (a) A typical coaxial connector. (b) Its formal representation 73
Figure 4.6 A network represented as a cloud 74
Figure 4.7 (a) A wound toroid. (b) Diagrammatic representation
of a wound toroid 75
Figure 4.8 A toroid with two windings and a toroidal screen between them 75
Figure 4.9 Diagrammatic representations of components 76
Figure 4.10 (a) An actual construction of an isolated source.
(b) Its schematic representation 77
Figure 4.11 (a) A voltage DU injected into an inner conductor.
(b) Detection of a current I in an inner conductor 77
Figure 5.1 A singleconductor AC Wheatstones bridge 80
Figure 5.2 A completely screened singleconductor AC Wheatstones bridge 80
Figure 5.3 A simplified line drawing of a completely screened
Wheatstone bridge 81
Figure 5.4 A pictorial representation of a simple capacitance ratio
bridge based on a voltage ratio transformer 82
Figure 5.5 A diagrammatic representation of the bridge represented
in Figure 5.4 82
Figure 5.6 The relationship between (a) a fourarm bridge and
(b) one based on a N:1 voltage ratio transformer 83
Figure 5.7 An equivalent circuit for a resistor 84
Figure 5.8 The direct impedance or admittance of a standard 87
Figure 5.9 A twoterminal definition of a component 87
Figure 5.10 A fourterminal definition of a component 88
List of figures xv
Figure 5.11 A coaxial fourterminal definition of a component 88
Figure 5.12 A twoterminalpair definition of a component 89
Figure 5.13 (a) The internal admittances of a twoterminalpair component.
(b) Simplified equivalent circuit 89
Figure 5.14 (a) The usual threeterminal definition of a component and
(b) its equivalent circuit 90
Figure 5.15 (a) The fourterminalpair definition of a component.
(b) Its equivalent circuit 91
Figure 5.16 A fourterminalpair component drawn as the sum
of two fourterminal components in series 91
Figure 5.17 A twoterminalpair bridge used to measure fourterminalpair
admittances 93
Figure 5.18 (a) An adaptor to convert two and fourterminal
components to a fourterminalpair definition. (b, c and d)
Rigid adaptors for particular applications 95
Figure 5.19 A cable added to a twoterminal component 96
Figure 5.20 A (a) fourterminal component with cables attached
(b) its equivalent circuit 97
Figure 5.21 Cables added to a twoterminalpair component 99
Figure 5.22 Converting a twoterminalpair standard to four terminal pair
by adding Tconnectors at its output ports 100
Figure 5.23 Cables added to a fourterminalpair component 101
Figure 5.24 Comparing twoterminalpair components 103
Figure 5.25 A DC Kelvin double bridge. The components within
the dotted box are a combining network 104
Figure 5.26 (a) An AC combining network and (b) its equivalent circuit 106
Figure 5.27 (a) Connecting two AC combining networks in parallel
to obtain an inphase and quadrature adjustment.
(b) The equivalent circuit 106
Figure 5.28 Connecting fourterminalpair impedances in series
using a combining network 107
Figure 6.1 The ThompsonLampard capacitance theorem 111
Figure 6.2 (a) A practical ThompsonLampard capacitor.
(b) The inwardfacing surfaces of four cylinders 113
Figure 6.3 (a) A parallelplate capacitor. (b) Parallel equivalent
circuit. (c) Series equivalent circuit 116
Figure 6.4 Frequency dependence of the magnitude of real and imaginary
components of a 1 pF parallelplate capacitor at frequencies
up to 100 kHz 119
Figure 6.5 (a) Calculated and estimated (from (6.12) and (6.13))
frequency dependence of the real and imaginary components
of a 1nF parallelplate capacitor having 0.05 mH series
inductance and D 10
4
rad at frequencies up to 1000 MHz.
(b) The corresponding difference between the calculated
and estimated values 122
xvi Coaxial electrical circuits for interferencefree measurements
Figure 6.6 A threeterminal capacitor and its diagrammatic representation 123
Figure 6.7 A Zickner capacitor 126
Figure 6.8 (a) Diagrammatic representation of a transformerenhanced
capacitance. (b) Its equivalent circuit 127
Figure 6.9 An Sclass standard resistor 131
Figure 6.10 Tnetworks to generate (a) highvalued direct
resistances or (b) lowvalued direct capacitances 132
Figure 6.11 Equivalent circuit of an inductive resistor 133
Figure 6.12 Equivalent circuit of a capacitive resistor 133
Figure 6.13 Compensating an inductive resistor 134
Figure 6.14 Compensating an inductive resistor over a wide frequency
range 134
Figure 6.15 Compensating a capacitive resistor 135
Figure 6.16 The equivalent direct impedance 135
Figure 6.17 A Campbell mutual inductor 136
Figure 6.18 The equivalent circuit of a mutual inductor 138
Figure 6.19 A Haddad coaxial resistance standard 141
Figure 6.20 The highfrequency 4TP calculable coaxial resistance
standard for broadband DC to 100 MHz operation
( IEEE 2005 [17]) 144
Figure 6.21 Equivalent circuit lumped parameter model of a bifilar
resistance standard 147
Figure 6.22 Gibbings 100 W reversed quadrifilar resistance standard whose
resistance increases quadratically with frequency and at 8 kHz
has a value 6.510
9
higher than DC (with t 16 ns) 148
Figure 6.23 (a) Highfrequency secondary 4TP resistance standard using a
(b) internal coaxial transmission lines (c) and surfacemount
chip resistor. The details (b) and (c) show the coaxial leads
before and after mounting the chip 151
Figure 6.24 A connection geometry for minimising parasitic
inductances of a chip component 152
Figure 6.25 A highfrequency fourterminalpair secondary
capacitance standard 153
Figure 6.26 The ideal design outline of the highfrequency 4TP calculable
coaxial capacitance standard for operation at frequencies
up to 100 MHz. The inset diagram shows a closer view
of the 4TP orthogonal connections ( IEEE 2007 [20]) 154
Figure 6.27 A highfrequency coaxial inductance standard of
calculable frequency dependence [20] 155
Figure 6.28 A possible design of a 500W frequencyindependent
(DC to 100 MHz) resistance standard 157
Figure 6.29 An ideal fourterminalpair coaxial resistance standard
of calculable frequency dependence 158
List of figures xvii
Figure 6.30 Calculated frequency dependence from 6.47 of the
ideal 1kW HF calculable resistance standard up to
100 MHz (bottom) and up to 100 GHz (top) 159
Figure 6.31 Calculated frequency dependence of the selfinductance of
the 1kW HF calculable resistance standard at frequencies up to
100 MHz (bottom) and 100 GHz (top). The erratic lowfrequency
behaviour is merely an artefact of the numerical calculations 160
Figure 6.32 The layered structure of a quantum Hall device 162
Figure 6.33 The terminals and layout of a quantum Hall bar 163
Figure 6.34 A photograph of a quantum Hall device mounted on a
printedcircuit board, which is 16 mm in diameter and
designed in the frame of the Euromet project No. 540 163
Figure 6.35 The quantum Hall and longitudinal resistance of a
GaAs heterostructure 164
Figure 6.36 A measuring chain from a calculable capacitor to the
DC quantum Hall resistance 166
Figure 6.37 A measuring chain from the DC quantum Hall
resistance to a 10pF capacitance standard 167
Figure 6.38 A measuring chain from an AC measurement of quantum
Hall resistance to a 10pF capacitance standard 168
Figure 6.39 The equivalent circuit of Ricketts and Kemeny and an
equivalent, simpler circuit 170
Figure 6.40 A tripleseries connection 173
Figure 6.41 (a) The relevant device edges of a sixterminal device.
(b) The equivalent circuit (for simplicity without the
longitudinal resistances) 174
Figure 6.42 The equivalent circuit of a quantum Hall device with
a tripleseries connection scheme (n m 3) including
the contact resistances and the lead impedances and
admittances. For simplicity, the outer conductors of the
coaxial leads are not shown 175
Figure 6.43 A shielded device holder 176
Figure 6.44 A device holder in a cryostat 177
Figure 6.45 A quantum Hall device mounted on a split backgate 182
Figure 6.46 A doubleshield device 183
Figure 7.1 The permeability of magnetic core material as a
function of flux density 190
Figure 7.2 Eddy currents induced by flux changes in (a) a solid
magnetic core and (b) a ribbonwound core 190
Figure 7.3 Flux leakage from a ribbonwound toroidal magnetic core 191
Figure 7.4 The effect of leakage flux on a transformer 192
Figure 7.5 The equivalent circuit of a transformer having leakage
flux and winding resistances 193
Figure 7.6 The internal capacitances of a transformer winding 193
xviii Coaxial electrical circuits for interferencefree measurements
Figure 7.7 The equivalent circuit of a transformer winding
including the effects of leakage inductance L, resistance R
and internal capacitance C 194
Figure 7.8 Connecting a small resistance between the windings 194
Figure 7.9 Equivalent circuit of a twowinding transformer 195
Figure 7.10 The equivalent circuit of a threewinding transformer 196
Figure 7.11 The equivalent circuit of the primary of a voltage transformer 197
Figure 7.12 A second core added to a voltage transformer 197
Figure 7.13 Parasitic impedances of a transformer having a
second core and winding 198
Figure 7.14 Avoiding the need for a separate auxiliary transformer 199
Figure 7.15 An uniformly wound toroid 200
Figure 7.16 A toroidal winding progressing around a core 200
Figure 7.17 A transformer winding with an antiprogression turn 201
Figure 7.18 A bootlace technique for transformer winding 201
Figure 7.19 Progression of a bootlace winding to minimise
selfcapacitance 202
Figure 7.20 A circuit for counting the number of turns of a
winding on a core 202
Figure 7.21 The ratio of numbers of turns measured using an AC voltmeter 203
Figure 7.22 A bifilar winding: (a) pictorial representation and
(b) diagrammatic representation 205
Figure 7.23 (a) Schematic representation of a bifilar winding.
(b) Capacitances between the conductors of a bifilar winding 205
Figure 7.24 (a) A rope winding on a transformer core. (b) A schematic
representation 206
Figure 7.25 (a) An ordered rope of conductors for a transformer
ratio winding. (b) A cross section 207
Figure 7.26 Testing the perfection of a nearly toroidal field 209
Figure 7.27 Connections to the output ports of a transformer 210
Figure 7.28 A simple IVD 211
Figure 7.29 Further subdivision of the input voltage of an IVD 211
Figure 7.30 Switched selection of the taps of an IVD 212
Figure 7.31 Minimising the number of cores needed for an IVD 212
Figure 7.32 (a) Twotap switching of an IVD. (b) The two final decades are
via a stepdown transformer. (c) Reducing the number of cores 215
Figure 7.33 Formal representation of (a) an IVD having a single
output and (b) an IVD having two independent outputs 216
Figure 7.34 The internal impedances of a singledecade IVD 217
Figure 7.35 The twostage principle applied to an IVD 217
Figure 7.36 (a) A manydecade twostaged IVD. (b) Its formal
representation 218
Figure 7.37 A pictorial section and a formal representation of an
injection or detection transformer 219
List of figures xix
Figure 7.38 Capacitive currents on either side of the shield
surrounding the inner conductor 219
Figure 7.39 An injection/detection transformer having an isolated winding
and its formal representation 220
Figure 7.40 A cross section through a detection transformer, which
detects or injects an unbalanced current in a coaxial cable
and its formal representation 220
Figure 7.41 An injection transformer as a small voltage source 221
Figure 7.42 A twostaged voltage injection transformer 221
Figure 7.43 Calibrating the inphase component of a voltage
injection transformer 222
Figure 7.44 Calibrating a voltage injection network.
(a) Coaxial circuit. (b) Principle of the circuit 223
Figure 7.45 Trimming the accuracy of a quadrature voltage injection 224
Figure 7.46 The construction of a twostaged voltage transformer 226
Figure 7.47 A current ratio transformer used to compare the ratio I
1
/I
2
of two currents 227
Figure 7.48 Winding impedances causing a ratio error in a current ratio
transformer 228
Figure 7.49 Calibrating an IVD with the output voltages of a fixedratio
transformer 229
Figure 7.50 The principle of calibrating a transformer 229
Figure 7.51 A schematic diagram of a practical circuit for calibrating a
voltage ratio transformer 231
Figure 7.52 A coaxial screened circuit for calibrating a voltage
ratio transformer 232
Figure 7.53 A straddling method of calibrating a voltage ratio
transformer 233
Figure 7.54 A scheme for calibrating a 10:1 voltage ratio 235
Figure 7.55 A section through a 1:1 calibrating transformer 235
Figure 7.56 A practical circuit to implement the straddling method for
calibrating a voltage ratio transformer equipped with defining
transformers J
2
and J
2
0
237
Figure 7.57 A multiport connector where inner conductors are brought
together at a point 238
Figure 7.58 A capacitance bridge to check the accuracy of the
straddling method 239
Figure 7.59 Principle of the permuting capacitors method for
calibrating a voltage ratio transformer 240
Figure 7.60 A practical circuit to implement the permuting
capacitors method 241
Figure 7.61 An arrangement of permuting capacitors suitable for
frequencies up to 10 MHz [5] 242
Figure 8.1 Shunting a fourterminalpair impedance 249
xx Coaxial electrical circuits for interferencefree measurements
Figure 8.2 (a) A twoterminalpair zero admittance.
(b) A twoterminalpair zero impedance 253
Figure 9.1 The principle of a twoterminal bridge based
on an IVD 256
Figure 9.2 A twoterminalpair bridge based on IVDs 257
Figure 9.3 A fourterminalpair bridge based on IVDs 259
Figure 9.4 A twoterminalpair bridge based on a 10:1
voltage ratio transformer 260
Figure 9.5 A fourterminalpair bridge based on a twostage
10:1 voltage ratio transformer 262
Figure 9.6 The principle of a quadrature bridge to relate
resistance and capacitance 263
Figure 9.7 Combining two quadrature bridges 264
Figure 9.8 The principle of a quadrature bridge coupled to a detector
combining network 265
Figure 9.9 The principle of a complete quadrature bridge 265
Figure 9.10 A twoterminalpair quadrature bridge 267
Figure 9.11 A fourterminalpair quadrature bridge 269
Figure 9.12 Principle of the MaxwellWien bridge 271
Figure 9.13 Threeterminalpair measurement of the connecting
impedance of an individual quantum Hall resistance lead 273
Figure 9.14 Equivalent circuit of a R
xx
measurement at the
lowpotential side of a quadrupleseries quantum Hall device 274
Figure 9.15 R
xx
measurement at the lowpotential side of a
quantum Hall device 276
Figure 9.16 Equivalent circuit of a R
xx
measurement at the
highpotential side of a quadrupleseries quantum
Hall device 277
Figure 9.17 Circuit diagram of the AC R
xxHi
bridge 278
Figure 9.18 A coaxial twoterminalpair ratio bridge for the
comparison of nondecade capacitances 281
Figure 9.19 Comparing a tripleseries connected quantum
Hall device, R
H1
, with a fourterminalpair
resistor, R
ref
282
Figure 9.20 Comparing two tripleseries connected quantum Hall
devices, R
H1
and R
H2
282
Figure 9.21 A quadrature bridge, which links two tripleseries
connected quantum Hall resistances, R
H1
and R
H2
, and
two capacitances, C
1
and C
2
, according to w
2
R
H1
R
H2
C
1
C
2
1 284
Figure 9.22 A quasifourterminalpair IVDbased bridge for
frequencies up to 1 MHz 285
Figure 9.23 Calibrating a 10:1 ratio transformer or IVD with a
permuting capacitors device 286
List of figures xxi
Figure 9.24 A complete 4TP 1MHz impedance bridge based on a
fixed 10:1 voltage ratio transformer and binary
IVD current sources 287
Figure 9.25 A bridge for comparing similar impedances 0.5 to 10 MHz 287
Figure 9.26 A 1:1 fourterminalpair 10 MHz resistance ratio bridge 289
Figure 9.27 A quasi fourterminalpair quadrature bridge
operating at 1.6 or 16 MHz 290
Figure 9.28 (a) A circuit for measuring the fourterminalpair
resonance frequency of a gasdielectric capacitor C.
(b) The fourterminalpair definition of the capacitor 292
Figure 9.29 An example of a connection between a fourterminalpair
impedance to a twoport vector network analyser.
The particular connection scheme shown is to measure
s
22
and s
42
294
Figure 9.30 (a) Scattering parameter measurements of a 1nF
capacitor. (b) The frequency dependence calculated
from these measurements 295
Figure 9.31 The autobalancing technique typically employed in
commercial instruments to measure a fourterminalpair
impedance Z 296
Figure 10.1 Schematic diagram of a CCC for comparing the
DC values of two resistors 301
Figure 10.2 Comparing voltage sources, for example, Zener standards 302
Figure 10.3 Comparing twoterminalpair resistances using two
AC independent programmable Josephson sources 304
Figure A1.1 TD transformations 307
xxii Coaxial electrical circuits for interferencefree measurements
Abstract
The central thesis of this book is that it is possible to design circuitry so that the
electric and magnetic fields associated with its currents and potentials are confined.
If this is done, an important consequence of reciprocity is that such circuitry does
not respond to external electric or magnetic fields it is immune to external elec
trical interference. Circuitry possessing these properties ought to be much more
widely used in electrical instrument, especially sensing and measurement systems
within the frequency range from DC to 100 MHz.
An electrical network of coaxial cables has no significant external magnetic field
if each outer conductor carries a current that is maintained equal and opposite to that
flowing in the inner conductor. If the outer conductors are all at nearly the same
potential, there will also be negligible electrical fields outside the cables. Conse
quently no cable has any means of interacting electromagnetically with any other
cable, and the performance of the circuitry also is unaffected by how the cables are
routed and is immune to interference.
This book describes examples of the elegance, power and utility of coaxial
networks. They should be used for precise and accurate sensing and measuring
circuits of all kinds. Moreover the principles involved can be applied to all circuits
to prevent them generating electromagnetic pollution. It updates an earlier pub
lication [1] by including more recent work such as AC measurements of quantum
Hall resistance to provide a primary quantum impedance standard and the extension
of coaxial networks to higher frequencies.
We apply first principles to common situations and then describe, as an
example of the ultimate in noise matching and interference elimination, the linear
networks for establishing and measuring standards of electrical impedance that are
known colloquially as coaxial AC bridges.
The authors have between them more than 60 years of experience in electrical
measurements carried out in national standards laboratories. These laboratories are
the source of measurement standards and techniques for the rest of the technical
and scientific world and are therefore dedicated to accuracy by reporting the unique
correct result of a measurement within a stated uncertainty. They continually strive
to reduce that uncertainty to stay ahead of the requirements of science and industry
for the foreseeable future and to promote good electrical measurement practice.
The principles elucidated in this book play a major part in this process.
Reference
1. Kibble B.P., Rayner G.H. Coaxial ACBridges. Bristol: Adam Hilger Ltd.; 1984.
(Presently available from NPL, Teddington, TW11 0LW, U.K. www.npl.co.uk)
Preface
Lord Kelvin is famously reputed to have said to measure is to know. This remark
might well be expanded, to measure is to know, to measure accurately is to under
stand. The process of accurate measurement starts with setting down complete
defining conditions for the measurement and devising measuring apparatus which
aims to fulfil these conditions. It continues by deliberately making changes in the
measurement method. If these changes still fulfil the defining conditions but different
results are obtained, basic principles cannot have been obeyed. Therefore, in order to
obtain a unique and accurate result, basic principles must be thoroughly understood,
and defining conditions derived from them obeyed by adopting this methodical
approach the very significant cost of accurate measurement will be minimised.
In particular, the basic principles of interference elimination are, in general, not
well understood. Everybody who designs, constructs or uses sensitive electrical mea
suring circuits has to contend with issues of extraneous electrical noise and inter
ference. A good system should be robust and insensitive to electrical and mechanical
disturbances of any kind. Often these disturbing influences are accompanied by
systematic errors.
The authors have gained considerable experience in combating the problem of
interference through many years of work on impedancemeasuring networks
(bridges). In this specialised activity, sensitivities of nanovolts or picoamperes are
common, and therefore great care has to be taken to prevent any part of the network
from coupling with any other, or with electromagnetic fields in the environment.
Prevention has been achieved through the elegant application of coaxial networks
where connections between screened components are made with coaxial cables in
which the currents in the inner conductor and the outer (screening) conductor are
constrained to be equal in magnitude but opposite in direction. The impedances to
be compared are completely defined in terms of the voltage differences and cur
rents to be established at their terminals. A network of this kind is electro
magnetically isolated, and its own electromagnetic fields are contained within its
components and interconnecting cables. Employing coaxial networks for accurate
impedance comparisons is an esoteric activity. Standardising national laboratories
obliged to acquire this capability may find that the equipment is not easily com
mercially available. They will be obliged to make it inhouse, but then there will at
least be the advantage of knowing intimately how it works. To assist this situation,
we give some information on constructional techniques.
Readers not concerned with accurate impedance standardisation and mea
surement will find these comparatively simple systems are nevertheless well worth
studying because they are an example of what constitutes good practice when
designing and constructing sensitive measuring circuits. We therefore make no
excuse for describing impedancecomparing circuits in some detail.
Some sections of the book (such as chapters 5 and 6) may only be relevant to
these more specialised readers. The book is also intended for those in national
standards laboratories who derive standards from the abstract definitions of SI units
with the greatest possible accuracy.
xxvi Coaxial electrical circuits for interferencefree measurements
Acknowledgements
The authors thank The PhysikalischTechnische Bundesanstalt for extensive sup
port and encouragement; and The National Physical Laboratory for permission to
include most of the content of NPL report DES 129.
We acknowledge the considerable contribution Dr. Ian A. Robinson has made
to our understanding of the science of eliminating electrical interference from sensi
tive circuits and the technical help of Norman and Charles Lloyd (N.L. Engineering,
Woodhurst, Huntingdon, Cambridgeshire, U.K.) in constructing many of the bridge
components described in chapters 69. Jurgen Melcher has derived the topological
result of section 3.2.2 and Appendix 2. We thank Dr. H. Bachmair for reading the
text and making valuable suggestions.
Shakil Awan would like to express special thanks and gratitude to Amber,
Gabriel and Liberty for their love, patience and support. Particular thanks to my
parents for their immense support, resilience and encouragement without which
parts of this book would not have been possible. Many thanks also to Sally and
David Luscombe for all their help and encouragement.
Miss Janet H. Belliss and Dr. Ian A. Robinson (NPL, U.K.) are gratefully
acknowledged for their help and support over the years. Thanks also to Dr. Luca
Callegaro (INRIM, Italy) for useful discussions and providing Figures 9.29 and 9.30.
Dr. Sze W. Chua (NMC/A*Star, Singapore) is also acknowledged for previous work
on quantum Hall impedance and for providing a diagram.
Bryan Kibble thanks Anne for her support, both in general and particularly
during the preparation of this book.
Chapter 1
Introduction
1.1 Interactions between circuits eliminating electrical
interference
Interference elimination is amenable to a logical approach and is most certainly not
a black art. It is necessary to first discard imprecise and often incorrect notions of
earthing and shielding before any real progress can be made. It is true that
signal processing can sometimes extract the required information from a hor
rendous cacophony of interference, but, nevertheless, the rewards of applying
correct, simple principles to produce clean, problemfree electrical circuitry are
very great.
For example, by adopting the methods described in this book to confine the
electric and magnetic fields associated with a circuit, whether DC or AC, to within
its components and cables, the expense and inconvenience of a screened room is
often unnecessary.
At the outset, we remark that it is wrong to think of a DC circuit only in DC
terms. It is also an AC network and, as such, is interference prone.
Avoiding unwanted interactions between circuits is important, not only with
regard to the subject matter of this book but also throughout the whole field of
sensitive electrical circuits where results need to be free from inaccuracy and
signaltonoise problems caused by power line pickup, interference from other
external sources and unwanted interactions between different parts of a system.
A system could be a network of impedances and instruments or the internal
workings of a single instrument. See Reference 1 for a practical example embody
ing the principles expounded in this book.
It is helpful to visualise the circulating currents and the magnetic fields that the
currents give rise to, the electric potentials of conductors and the electric fields
emanating from the conductors as the physically real entities. Unfortunately, the
vast majority of electrical test and measuring instruments used to explore condi
tions in a circuit (e.g., voltmeters, oscilloscopes and spectrum analysers) are
regarded as responding to input voltages. It must be remembered at all times that
these voltages are only manifestations of currents flowing through impedances
between two points in their circulating paths.
1.1.1 Basic principles
CURRENTS FLOW AROUND COMPLETE CIRCUITS
Currents start from a source for example, a battery, power supply, signal gen
erator, a voltage generated between the ends of a conductor in which another cur
rent is flowing but having traversed a possibly very complicated network of
conductors and impedances, subdividing and recombining on the way, they must
recombine and return in their entirety to that source. They do not go to ground,
earth or a shield and disappear! For example, inductance can only be properly
understood if it is realised that it is only defined for a closed circuit. What is usually
meant by a statement about inductance is incremental inductance, such as that
caused by a length of wire if it were to be inserted into a closed circuit.
The concepts of grounding, earthing and shielding, which are of historical
origin, have caused much confusion and misery to practitioners of sensitive electrical
measurement, mainly because of the woolly manner in which these concepts are fre
quently invoked. We suggest as an alternative that one point of a network be desig
nated as a potential reference point to which all other potentials in the network are
related and which is therefore usually given the value of zero potential. It will usually
be about the same potential as the earth or surroundings of the apparatus walls,
water pipes, people, etc. bearing in mind that their potentials are rather uncertain.
It follows that the actual earthing or grounding conductors, which are repre
sented by symbols like those in Figure 4.9n, have flowing in themcurrents returning to
their source. Except for the very special case of superconductors with DC, these con
ductors must have potential differences along them. This can most certainly matter in
sensitive circuits. The actual return conductors and the precise points at which they are
interconnected should be drawn and considered together with the currents flowing in
them. Particular care should be taken to represent the topology of the connections of
the actual circuit correctly. For an example, see Figure 1.14. It is also helpful to replace
the words earth or ground in our vocabulary with more specific terms such as
current return conductor, instrument case and mains safety conductor as appro
priate, and to designate a particular node of a network as a potential reference point to
which all voltages are understood to be referred. The time spent thinking in this more
precise way will be amply repaid, particularly in eliminating commonimpedance
coupling described below. Amains safety conductor is often coloured green/yellow. It
is the only meaning of the Figure 4.9n symbol in our diagrams.
Currents that flow around complete circuits give rise to electric and magnetic
fields, which cause interaction between circuits and between different parts of the
same circuit.
The reciprocity theorem is embedded in the theory of all fields, not just elec
tromagnetic. It connects the emanation of fields by a system to the response of that
system to external fields. For our purpose it is sufficient to make use of a special case.
CIRCUITRY WHOSE LAYOUT AND SCREENING ARE SO DESIGNED
THAT IT EMITS NO SIGNIFICANT ELECTRIC OR MAGNETIC FIELDS
DOES NOT RESPOND TO EXTERNAL FIELDS. THEREFORE, IT IS
IMMUNE TO INTERFERENCE.
2 Coaxial electrical circuits for interferencefree measurements
We can classify these modes of interaction as follows:
(i) Electric field coupling: Either two capacitances or, more usually, one capa
citance and a linking conductor (which is frequently the green/yellow or
earth conductor to an instrument) can couple two systems via their external
electric fields, as shown in Figure 1.1 or Figure 1.2.
Electric field coupling is more important in highimpedance circuits
having typical impedances of 10 kW or greater.
Capacitive coupling can be eliminated by shielding with conducting
surfaces connected to appropriate points in the circuitry. Shields can inter
cept the capacitive currents and confine them to the interfering source, as
shown in Figure 1.3, or can prevent these currents from flowing through
sensitive parts of the detecting circuit.
Many instruments and components may appear to be encased in fairly
complete conducting enclosures, but this can be illusory if the panels of these
enclosures are, in fact, insulated from one another by, for example, an oxide
D
Figure 1.1 Coupling via two capacitances
D
Figure 1.2 Coupling via one capacitance and a conductor
D
Figure 1.3 Placing a screen to eliminate capacitive coupling
Introduction 3
layer or paint. Also, the shafts that mechanically couple internal controls
such as switches or potentiometers to knobs or buttons outside the case are
not always connected to the case and therefore provide a means of allowing
currents via electric fields to enter or escape (Figure 1.4a). In Figure 1.4b, the
offending shaft is connected to the case, and the effect of an oxide or paint
layer is minimised by joining input and output connectors with a conductor
routed close to the circuitry within the enclosure. The enclosure should then
have a deliberate insulating gap so that the added conductor provides the only
path for returning current. Doing this can minimise the open loop area exposed
to external interfering flux, as discussed in (ii) below.
Some aspects of capacitive coupling between various parts of a circuit,
for example, between conductors on the surface of a printed circuit board,
are often misunderstood. The basic principles of this situation are illustrated
in Figure 1.5.
1 2 3
C
12
C
13
C
23
Figure 1.5 Capacitances between conductors on a printed circuit board
The capacitance C
12
between conducting surfaces 1 and 2 is modified by
the presence of a third conducting surface 3. If 3 is maintained at a fixed
potential with respect to 1 and 2, C
12
is decreased somewhat to, say, C
0
12
. This
change is independent of the particular potential of 3, be it zero (ground)
potential or several kilovolts with respect to 1, 2 or any other nearby conducting
surfaces. If the potential of 3 is indeterminate that is, if it is floating, that is,
not connected to circuitry which determines the potential of 1 and 2 the current
through C
12
will be augmented by that through C
13
and C
23
in series.
We can apply these concepts to two conductors on a printed circuit board,
which is provided with a conducting ground plane in the intervening
space between them, as shown inFigure 1.5, or covering the whole of the reverse
side of the board, as shown in Figure 1.6. In either case, the direct capacitance
Z
I
Z
I
I
I
I
I
good not good
spring
I I
(b) (a)
Figure 1.4 (a) A conducting control shaft penetrating a shield. (b) Connecting
the shaft to the shield with a spring contact
4 Coaxial electrical circuits for interferencefree measurements
C
0
12
between the two conductors is only approximately halved. In a circuit con
taining only impedances that are low compared with 1/(2pfC
13
), etc., the
ground plane has no capacitive role, but at higher frequencies ( f ) when its
impedances 1/(2pfC
13
) etc. are lower, the potential divider formed by C
0
12
and
either C
23
or C
13
may well be significant and useful. For more complete
screening, there shouldbe a conductingplane above, belowandbetween 1 and 2.
(ii) Complete meshes of networks can be coupled via the magnetic field arising
from current flow around one of them (Figure 1.7). That is, circuits can be
coupled by mutual inductance. This coupling can be reduced, as shown in
Figure 1.8, by reducing the effective area of one or, preferably, both of the
meshes by, for example, using twisted wire conductor pairs or the coaxial
cables, which are the subject of this book. This kind of coupling dominates in
circuitry having impedances of less than 10 kW.
D
I
B
Figure 1.7 Inductive (magnetic) coupling
D
I
Figure 1.8 Eliminating inductive coupling
It is a fundamental property of a straight, infinitely long coaxial cable in
which the current in the inner conductor is balanced by an equal and opposite
current in the outer screening conductor that these currents generate no mag
netic flux external to the cable (Figure 1.9). In this book we mean by coaxial
cable a cable having the currents in the inner and outer conductors equal and
opposite. From Amperes theorem, since there is no net current in the cable, the
line integral of the flux around any circular path around the axis of the cable is
zero, and since, by symmetry, the flux at every point along the path must be the
same, this flux must be zero everywhere outside the cable.
3
C
12
C
13
C
23
2 1
Figure 1.6 A ground plane on the reverse side of a printed circuit board
Introduction 5
I
I'
B l I I' ) = 0 d = (
0
B l d
C
C
Figure 1.9 Amperes theorem applied to a circular path concentric with the axis
of a coaxial cable
In practice, actual cables are neither infinite nor straight, but if the
length and radii of curvature of bends are long compared to the diameter of
the cable, the external flux will be negligible.
The same is true for twistedpair cables in which the currents are equal
and opposite because the flux generated by successive twists cancels at dis
tances larger than the pitch of the twisting (see Figure 3.10).
Components can, if necessary, be contained within enclosures of highly
permeable magnetic material such as mumetal. An unfamiliar example,
which will be encountered later in the book (see section 7.2.6), is the enclosure
of a toroidal transformer core, drawn in section in Figure 1.10a, in a toroidal
magnetic shield. The purpose of doing this is to ensure that the magnetic flux
associated with the core windings is completely confined to the interior of the
shield. This technique assists in the generation of precise voltage ratios by
windings threading the toroidal shield. Figure 1.10b illustrates a practical way
of constructing a toroidal magnetic shield.
(iii) Figure 1.11 illustrates two circuits interacting because they share a length of
conductor. The current I circulating around the righthand circuit sets up a
voltage between the ends of the common conductor, which in turn causes a
current i to flow around the lefthand circuit. This commonconductor cou
pling can be equally important in both high and lowimpedance circuitry. In
addition to this coupling mode there can be significant contributions from
modes (i) and (ii), that is, via stray capacitance and magnetic flux linkage.
I
I
U
I
(a)
insulating sleeve metal halfshell metal halfshell
(b)
Figure 1.10 (a) A toroidal magnetic shield surrounding a toroidal magnetic core.
(b) A practical construction for the shield
6 Coaxial electrical circuits for interferencefree measurements
D
Z I i
Figure 1.11 Commonconductor coupling
Assuming that there is a reason for connecting the two circuits, pure common
conductor coupling can be eliminated either by reducing the common impedance to
zero by joining the two circuits with a star connection, as shown in Figure 1.12, or a
branch connection, as shown in Figure 1.13. A branch connection makes it easier to
also eliminate capacitance coupling with a shield, as in (i), and flux coupling, as in (ii).
Another possibility is a combining network, as discussed in section 5.6.
As an example of eliminating commonconductor coupling, consider simple
operational amplifier circuits. These could well be the first stage of matching and
amplification after a sensor before conversion to a digital signal. For this purpose,
we assume ideal operational amplifiers, which, when external feedback compo
nents are added, maintain U
in
0 between their input terminals with negligible
current flow into either and which ensure that neither input nor output is influenced
significantly by the magnitude of either supply voltage, U
+
or U . Actual opera
tional amplifiers approach this ideal rather closely.
A common application is the unsymmetrical amplifier circuit drawn in
Figure 1.14. I is the total feedback current through R
f
, and i
0
is the total output current
flowing through the load R
L
. Because of the high voltage gain of the operational
amplifier, the feedback operates such that U
in
&0 at all times. Therefore,
I
U
0
R
f
1:1
D
Figure 1.12 A star connection to eliminate commonconductor coupling
D
Figure 1.13 A branch connection to eliminate commonconductor coupling
Introduction 7
and
U IR
in
0 1:2
The voltage gain is
U
0
U
R
f
R
in
1:3
Also,
i
0
U
0
R
L
R
f
R
L
I 1:4
Usually i
0
is considerably greater than I. As an example, for an amplifier
having a voltage gain of 100, we might have R
f
10
5
W, R
in
10
3
Wand R
L
10
3
W,
so that i
0
100I. If U 10 mV, I 10 mA and i
0
1 mA. Note that I and i
0
complete their circuits through the power supplies, which supply fractions a and b,
respectively, of the total current, plus the currents (not shown) needed to operate
the internal circuitry of the amplifier. Because i
0
and therefore the output voltage
across the load have been made immune, by the design of the operational
amplifier, to variations in either supply voltage, it is highly immune to noise
voltages in either supply, and the noise current i
n
from any such noise voltage
flows only around the supply loop.
This simple situation becomes much more complicated when two amplifier
circuits are connected in cascade and fed, as is usual, from the same supply rails.
The circuit is drawn in Figure 1.15.
Commonconductor coupling at C provides additional input to the second
operational amplifier, and if the operational amplifiers are, in practice, not com
pletely immune to changes in the supply voltages, commonconductor coupling
along the supply rails and current return conductor can cause problems. In severe
cases, these commonconductor couplings can cause feedback and oscillations in
highgain circuits.
Some of these problems are avoided if the circuit connections are remade as
shown in Figure 1.16. Star connections avoid some commonconductor couplings,
I
U
R
in
R
f
R
L
I
I
i 0
i 0
U
+
U
U
in
U
0
i
0
i
0
+ I
b(i
0
+ I )
a(i
0
+ I )
Figure 1.14 The currents flowing in an unsymmetrical amplifier circuit
8 Coaxial electrical circuits for interferencefree measurements
and the differentialinput properties of both operational amplifiers sense the remote
voltage source. Mutual inductive coupling (see (ii) earlier) should be minimised.
The whole subject of interference elimination is often viewed as something of
a black art with ad hoc cures such as ferrite beads or mains interference filters
being used on a trialanderror basis. One cause of the adoption of this straw
grasping approach lies in the vector addition of induced voltages A, B, C, etc.,
which come from a common primary source. For example, consider each single
frequency component of mains and mainsborne interference entering circuitry by
the routes identified earlier in this section. The voltage induced at each frequency
by each route will have a different magnitude and phase associated with it, and the
resultant is the vector sum S, as illustrated in Figure 1.17. S is observed between
two points of the circuit by some test instrument such as an oscilloscope or volt
meter, and as a further complication, some additional voltage components may
arise from the loops added to the network by the very act of connecting the test
instrument.
Suppose that one suspected cause of entry of interference is identified and
removed so that one of the contributing voltage vectors, say C, is eliminated. The
remaining vectors, represented by the broken lines, then sumto give a newresultant S
0
,
A
B
U
R
f
R
f
'
R
L
'
R
L
R
in
Figure 1.16 Removing commonconductor couplings from the circuit of
Figure 1.15
I
I
I
U
C
R
in
R
f
R
f
'
a(i
0
+ I ) A(I
0
+ i
0
) a(i
0
+ I )
B(I
0
+ i
0
) b(i
0
+ I ) b(i
0
+ I )
R
L
R
L
'
i
0
i
0
I
0
I
0
I
U
0
0
Figure 1.15 The currents in a twostage operational amplifier circuit. Heavy lines
indicate occurrences of commonconductor coupling
Introduction 9
which will very possibly have a greater magnitude than S. There is now a strong
tendency to think that the action taken to eliminate C was incorrect; to restore the
status quo, abandon rational thought and try some other remedy at random instead.
In these circumstances, one must revert to sound basic principles and have the
courage of ones convictions to persist with correct remedial steps, even though
some appear to make the total observed interference worse. When all causes are
dealt with, the interference will disappear totally. To stop short of this ideal is to
court disaster because a small resultant can be obtained by the accidental partial
cancellation of many large contributions. Indeed, one terrible approach to elim
inating interference is to deliberately introduce a further large component equal and
opposite to S. The result will only be temporarily satisfactory because any change
in the circuit layout, the nature of the interference and so on could alter the mag
nitude or phase of one or more of the vectors and bring back the problem, probably
unsuspectedly.
The good news in this difficult situation is that if one frequency component, for
example, the fundamental harmonic of 50Hz mains interference, is properly
eliminated, all other frequency components including those of mainsborne inter
ference from other sources will also be eliminated because the underlying causes,
commonconductor, capacitive and mutual inductive coupling, have been reduced
to insignificance.
In the literature on interference rejection, reference is often made to problems
caused by earth loops. This phrase usually means the accidental or deliberate
provision of more than one current return path in the interconnections between
circuits. Quite large interfering voltages can be propagated amongst these circuits
by commonconductor or mutual inductive coupling. These voltages can be
eliminated by removing unnecessary return conductors and making appropriate star
connections between the remainder.
Finally, by applying the reciprocity theorem to balancedcurrent coaxial or
twistedpair networks, which generate no significant external magnetic or electric
fields, we reiterate the vital conclusion from this section that these networks do not
respond to external fields. If commonconductor coupling has also been eliminated,
these networks are interference free.
A
B
C
D
D
S
S'
U
X
U
Y
Figure 1.17 S is the vector sum of several (A, B, C, etc.) phaserelated
interferences from a common source. S
0
is the new vector sum if C is
eliminated
10 Coaxial electrical circuits for interferencefree measurements
1.1.2 An illustrative example using a phasesensitive detector
Consider a phasesensitive detector used as a null detector to detect the absence of a
signal. Measuring a wanted signal involves exactly the same considerations in that
the object is to ensure that there is no unwanted component added to this signal.
Phasesensitive detectors most acutely exemplify this problem because they need
an auxiliary reference input, which must be phaselocked to the signal to be
detected. Incorrect wiring can cause coupling of this reference signal into the
measurement input, and this causes consequent erroneous results, particularly
where a signal source is used to both stimulate the measurement circuitry and
provide the reference signal, as illustrated in Figure 1.18. Currents at the source
frequency circulate in the outer conductors of coaxial cables, which link the source,
measurement circuit and detector, and this results in mutual inductive and
commonconductor coupling in the detector. The important point is that there
should be no significant current in the inner or outer conductor of the coaxial cable
connecting the measurement circuitry to the signal input of the detector. This is
because a null indication of a properly constructed detector means that the voltage
difference across its input port is zero and that there is no current in the inner
conductor. Therefore, a necessary and sufficient condition for an accurate null
indication, when connected by a cable to a remote port in the measuring system, is
that there should be no net current in this cable. Current equalisation, by incor
porating a current equaliser (see section 3.1.1) into each mesh of the network, can
achieve this condition. A simple test for the effect of any current in the outer
conductor is to temporarily insert a resistance in series with the outer conductor.
Doing this, if all is well, should have no effect on the indication of the detector.
An alternative approach is to use isolating transformers. The relevant parts of
the network, including the green/yellow safety conductor of the mains, which also
connects sources and detectors, are shown in Figure 1.19. There are two undesired
currents. First, if the two screens of the isolating transformer A (see section 1.1.5)
are not quite at the same potential, a current will flow across the capacitance
between them, through the network of outer conductors in the bridge, through the
outer of the cable to the detector input and return to the transformer primary and its
measuring
system
phasesensitive
detector
modulation
source
reference signal
PC
electrical
data bus
Figure 1.18 Connecting a phasesensitive detector to a network
Introduction 11
screen via the green/yellow mains safety conductor and the reference supply cable
outer conductor (see Figure 1.19). Second, a proportion of the reference signal
current will return to the source via the green/yellow mains safety conductor and a
small proportion via the detector output lead outer conductor, the bridge output
outer conductors and the interscreen capacitance of the transformer. The currents
from both of these causes flow down the outer conductor, but not the inner, of the
input lead to the detector and cause a potential drop down the outer conductor.
Hence, the condition U 0 between the inner and outer conductors, which holds at
the detector input when it registers a null, does not hold where this input cable joins
the network, and this leads to an incorrect indication.
The net current in the detector lead can be detected by the total current
detection transformer described in section 7.3.5 and may be reduced to an accep
tably low value in either of the following two ways.
(i) An isolating transformer B in the reference signal cable will prevent current
flowing in the outer conductor of this cable.
(ii) An additional single conductor may be used to join the secondary screen of
transformer A to the outer of the detector case, and an equaliser put in the
cable to the detector input (Figure 1.20). Alternatively, an auxiliary small
voltage source comprising an injection transformer and associated circuit (see
section 7.3.4) may be connected between the primary supply outer and the
primary screen to modify the potential of the latter until it is equal to that of
the secondary screen, so that no current flows between them, as registered by
total current detection transformer surrounding the detector input lead.
1.1.3 Diagnostic equipment
There are some devices that, when used in conjunction with an oscilloscope or
spectrum analyser of the requisite bandwidth and sensitivity, are particularly useful
in revealing the causes of problems with interference.
measuring
system
modulation
source
reference signal
PC
optical fibre
databus
phasesensitive
detector
A
B
Figure 1.19 An isolating transformer can break the loop
12 Coaxial electrical circuits for interferencefree measurements
The ubiquitous handheld multimeter is a very useful diagnostic tool whose
limited accuracy of the order of 0.1% is usually adequate. Being battery powered
confers on it a fair degree of isolation, but nevertheless, simple precautions are
necessary in order not to get misleading results. The metal tips of probes should not
be held in the hands. Apart from the obvious safety implications, the shunt leakage
currents through the human body will cause errors when measuring high resistances
or voltages of sources having high internal impedance. When measuring alternating
voltages having associated high internal impedance, capacitance between the leads
or via the operators body can be significant, and separating the leads from each other
and placing the instrument on an insulating surface might reveal the existence of this
effect. Conversely, when sensing alternating voltages across very low impedances,
the inductance of the loop of separated leads could be significant and can be reduced
by twisting the leads together.
A batterypowered oscilloscope having a wide bandwidth is an extremely
useful diagnostic tool for eliminating interference. Its degree of isolation is very
helpful in not altering circuit conditions by the very act of connecting it into the
circuit. Mainspowered oscilloscopes with effective differential inputs can also
be useful, but the consequences of connecting the mains safety conductor to the
circuit being examined need consideration. A spectrum analyser is equally useful,
but these instruments are not usually completely isolated, and again, this may have
consequences.
A differentialinput preamplifier, to a large extent, avoids the problem of
additional significant current paths with concomitant interference being introduced
by the very act of connecting the diagnostic equipment itself. A properly designed
differential input is a threeterminal device, as illustrated in Figure 1.21, which has
r
e
f
e
r
e
n
c
e
s
i
g
n
a
l
D
optical fibre data bus
PC
measurement
network
Figure 1.20 An added conductor and an equaliser reduce the current in the outer
conductor of the input cable to the detector
Introduction 13
the property of responding to a voltage input between terminals 1 and 2 while
having a very much smaller (commonmode) response to voltages between 1 or 2
and 3. Therefore, there will be a true indication of circuit conditions if terminals
1 and 2 are connected to the nodes of the circuit between which we want to observe
the voltage. Careful thought must be given, however, as to where to connect
terminal 3. If there is no other connection between the circuit and the test instru
ment, via, for example, the green/yellow safety conductor, terminal 3 should be
connected to a point on the current return path of the circuit. If there is already a
lowimpedance connection of this kind between the circuit and the preamplifier,
terminal 3 should be left unconnected to avoid introducing an extra current return
path. The existence or otherwise of a connection can quickly be determined by a
handheld resistancemeasuring meter connected between a point on the current
return path of the circuit and terminal 3.
A sensitive differentialinput preamplifier correctly used in this manner can
detect commonconductor problems by connecting inputs 1 and 2 to the two points
on the conductor. An observed voltage indicates a commonconductor current in
the conductor, which may cause a problem.
We have emphasised that it is best to view the currents in circuitry as having
primary physical significance, but that, unfortunately, detection and measurement
of voltage differences are almost universal. But if it is too difficult to unravel the
problems of troublesome circuitry by making voltage difference measurements
alone, direct detection and measurement of currents can prove very enlightening.
This is possible if the conductor carrying the alternating current can be threaded
through a highpermeability magnetic core, which is provided with a winding
connected to a voltagedetecting instrument. This operation can be made very easy
if the core can be temporarily split so that it can be clipped around the conductor.
Direct currents can also be detected in the same way if the core has a transverse gap
in which a Hall effect sensor of magnetic flux is mounted. Clipon ammeters are
familiar instruments based on this principle, and by using a highpermeability core
and a multiturn winding, sensitivity can be extended up to the milliampere or even
the microampere range. The diagnostic power of this approach in difficult situa
tions cannot be overemphasised, but unfortunately, suitable commercial equipment
is rare.
A short solenoidal coil wound on an insulating former and covered with a
conducting screen, as shown in Figure 1.22, can be connected to an oscilloscope.
The screen should have an insulated overlap so that it does not form a shorted turn.
1
2
3
Figure 1.21 A differentialinput preamplifier
14 Coaxial electrical circuits for interferencefree measurements
The output of the coil is an induced voltage proportional to any AC magnetic flux
threading the coil. An induced voltage reveals the presence of unwanted magnetic
flux threading the circuit under examination. Employing equalisedcurrent circuits
and enclosing components in magnetic screens will eliminate problems caused by
magnetic flux.
If the coil is deliberately supplied with current from a source in the circuit, its
magnetic flux can test the sensitivity of the circuit to other unwanted flux couplings
by positioning the coil at various positions adjacent to the circuit. This can be a
very sensitive test to be interpreted with care as the flux the coil produces is likely
to be much greater than the actual unwanted fluxes present.
A capacitive probe is simply a sheet of conductor, preferably covered with an
insulating layer to protect circuitry from shortcircuits caused by accidental contact,
which is connected via a length of screened cable to the input of an oscilloscope or
some other detecting instrument (Figure 1.23). When held by the screened cable (to
connect the handler to the screen) and placed in the vicinity of the circuit being
examined, it will reveal the presence of any unwanted electric fields, which might
be part of a capacitive current flow. In a similar manner to the magnetic pickup
coil, it can be connected to a generator to create electric fields to which circuitry
might be sensitive. Because of the highimpedance nature of this useful device, it is
somewhat harder to interpret the results in comparison with those obtained with a
magnetic pickup coil.
E
D
detector or
generator
D
E
Figure 1.23 Probes to detect or produce electric fields
Coaxial networks can be used over a very wide frequency range, from DC to
several GHz. In principle, the frequency range over which coaxial cables not
terminated by their characteristic impedance operate in networks is limited only
by the requirement to keep the length of the cables rather less than the wavelength
of the electromagnetic field in the cable. In some applications, the phase change
down a cable, which occurs if its length becomes comparable with the propagation
B
D
detector or
generator
Figure 1.22 A pickup coil for detecting magnetically coupled interference
Introduction 15
wavelength in the cable, is a serious drawback. To overcome this, cables can be
kept shorter if the components of the network are physically smaller. Typical
figures are less than 1 m at 1 MHz, 100 mm at 10 MHz and so on. This need for
increasing miniaturisation will no doubt reach the ultimate in micromachined cir
cuitry as this technology develops.
For planar circuits on circuit boards, at still higher frequencies, a continuous
sheet of conductor on one side of the board, termed a ground plane, is often
employed. Currents in circuitry just above this plane prefer to flow in routes of
minimum inductance by creating return paths in the plane immediately below
themselves. There may still remain capacitive couplings between parts of the
circuitry above the plane.
The concept of a linear network is very powerful in designing and under
standing networks of linear circuit elements, that is, those for which the voltage
across them is strictly proportional to the current through them under normal
operating conditions.
The superposition theorem provides an extremely powerful tool for unravelling
complicated network problems. It should be borne in mind at all times when con
sidering interference elimination. It states that in a network of linear circuit
elements for example, impedances, subcircuits and sources in which the vol
tage across them is proportional to the current through them, the current from each
individual source behaves as if no other currents are present. We can conceive of a
current setting out from a source, dividing and subdividing as it goes into the
network, and then these separate components of the current combine and recom
bine with the original current value just before returning to the source. Super
position is much more than an abstraction it has immediate and practical
application in many circumstances.
In a linear network, many currents, each existing independently and originat
ing in separate sources, can sum to give the total current in any one path of
the network.
A source can be the voltage drop across a component. If one were to replace
the component by a voltage source equal to the voltage across it, the rest of the
network would have no means of registering the replacement.
There are many theorems relating to linear networks Kirchoffs laws,
Thevenins theorem, etc. but many useful conclusions can be reached directly
from the above basic principles without recourse to mathematics.
But beware if some circuit components are not sufficiently linear, as, for
example, semiconductor junctions and poor joints. The result will be generation of
harmonics from an original sinusoidal source and, in general, voltagedependent
responses.
The reciprocity principle is also vital for understanding some aspects of cir
cuitry. It is deeply embedded in the theory of classical fields, including electro
magnetic ones. As a simple example, there is reciprocity of source and detector in a
Wheatstones bridge network. That is, if the bridge is balanced so that the current
through the detector is zero, on interchanging the positions of detector and source,
the detector current is also zero that is, the bridge is still balanced. On a more
16 Coaxial electrical circuits for interferencefree measurements
sophisticated level, the directional radiation pattern of a transmitting antenna is
identical with the sensitivity receiving pattern of the same antenna.
For our present purpose, we have noted that we can more easily visualise how
a circuit will emit electromagnetic fields than how it will respond to external ones.
By arranging its layout and screening, we can minimise its emission and therefore
ensure that it responds less to external interference. Coaxial techniques, properly
applied, can reduce sensitivity to interference to negligible proportions. The most
powerful tool of all in the battle with interference is to develop the imaginative
powers of the brain for visualising the electric and magnetic fields generated by
circuitry and then take steps to minimise these.
Interference is usually, but not invariably, conveyed electrically. Earlier we
have assumed that all interference coupling modes between circuits are electrical.
In fact, significant energy can sometimes be coupled acoustically that is, the
interfering circuit can have some element that behaves like a loudspeaker, and
the measuring circuit, an element that behaves like a microphone. Examples of
the former include mechanical movement of power transformer cores and con
ductors, vibration of cooling fans, piezoelectric elements and electric field or
mechanically induced movement of shielding conductors. Examples of the latter
are acoustic modulation of the reluctance of matching and isolation transformer
cores, triboelectric effects in dielectric insulators and capacitance modulation of
shields. Coupling modes can often be identified simply by altering the damping
of the emitter, transmitting medium or receiver by clamping or grasping in the
hand, or by altering the disposition of components. Once identified, it is usually a
simple matter to provide the necessary acoustic isolation by using nonmicrophonic
cables for critical connections and acoustically lossy mountings for troublesome
components.
To summarise, eliminating electrical interference depends on applying ele
mentary principles concerning circuit coupling. Some persistence is often neces
sary in real, complicated situations to avoid ad hoc unsatisfactory partial solutions.
1.1.4 Isolation
Eliminating all possible sources of extraneous interference is vital for successful
sensitive analogue electrical measurements. If interference is present, not only is
the resolution of the measurement impaired through degradation of the signalto
noise ratio but also the rectification of the interference by any of the myriad
semiconductor rectifying junctions in modern electrical circuitry may well result in
systematic error. The ideal, to be attained as closely as possible, is to isolate the
lowlevel circuitry. That is, to achieve that no significant extraneous currents enter
or leave the isolated circuits.
When the necessary isolation has been accomplished, there remains the pro
blem of conveying of the analogue information gained by the lowlevel circuitry
from the isolated region, either in analogue form or, more commonly and satis
factorily these days, in digital form, without destroying the integrity of the
isolation. We discuss the principal ways of accomplishing this.
Introduction 17
Once in highlevel analogue or in digital form, the information should be
reasonably safe from further significant corruption by interference. This is parti
cularly true for on/off pulsed digital information, on which interference below a
certain maximum level has no effect at all. The information can then be safely
further processed by computers.
By proper design, it is possible to totally isolate a circuit so that it can only
communicate electrically with anything outside its screened boundary by desired
couplings. This concept, which is central to interference rejection, can be appreciated
by referring to Figure 1.24. The processing circuit may well contain one or more
noise sources, and there must be a filter to short the output of these sources to its local
screen so that the noise currents are constrained to follow paths that are totally within
the local screen. The circuit should be constructed by either using the conductorpair
techniques of section 1.1.10 or making the local screen so that magnetic or electric
fields do not escape it. The local screen is connected to the overall screen, which
isolates the measurement system as a whole from external electric fields, at just one
point A, as shown. The measurement shield may be connected to the green/yellow
mains safety conductor for safety reasons at just one point B.
filter
circuit
containing
sources of
interference
measurement
circuit
A
B
mains safety conductor
Figure 1.24 An isolating enclosure with a filtered output
The two examples in section 1.1.5 and 1.1.6 should help to clarify how to
achieve isolation in practice.
1.1.5 Totally isolating transformers and power supplies
Consider the mains power supply circuit outlined in Figure 1.25.
The filter consists of storage capacitors and voltage regulation circuitry.
These ensure that only a DC potential is allowed to exist between the output
terminals. The transformer is uniformly wound on a toroidal magnetic core, and the
winding has an antiprogression turn (see section 7.2.1) to cancel the singleturn
effect of the winding as it advances around the core. This form of transformer
winding does not significantly propagate magnetic flux into its surroundings.
The two toroidal shields over the primary winding need to be constructed with
their circumferential overlaps as shown in Figure 1.26, which shows a cross section
through the axis of a toroidal core and these nested shields. For perfect capacitive
18 Coaxial electrical circuits for interferencefree measurements
isolation, the crosscapacitances between them at their overlapping gaps, which are
of the order of a few picofarads, need to be equal, and they are best made as shown
in Figure 1.27d. If necessary, this can be done by connecting a trimmer capacitor in
parallel with the lesser crosscapacitance. This adjustment can be carried out by
connecting a highly sensitive oscilloscope between the inner and outer shields and
adjusting the capacitor for minimum observed 50Hz signal.
It is also possible to construct screens of a poorly conducting material such as a
loaded plastic. The conductivity of the sheet material should be sufficient to act as a
screen for electric fields but low enough so that a complete screen having a con
ducting joint rather than an overlap does not constitute a significant shorted turn of
the transformer.
rectifying,
regulating
and filtering
circuitry
measuring
circuit
A
B
mains safety
conductor
Figure 1.25 A totally isolated mainsdriven power supply
mains
input
secondary
output
core
Figure 1.26 A cross section through a toroidal transformer for a totally isolated
supply
Introduction 19
good
(a) (b)
(c) (d)
not good
not good
not good
U = 0
U
0
U
0
0 U
U U
0 0
0
U
0
U
0
U U
0
Figure 1.27 The best way to overlap nested toroidal shields
The isolation of a supply constructed along these lines can be such that only a
fraction of a microampere of current will flow at mains frequencies from the pri
mary to the measurement system, and thence to the measurement shield and back to
the primary winding.
1.1.6 Isolating a noisy instrument
Many commercial instruments, for example, digital voltmeters, are insufficiently
isolated from the mains by their internal power supplies. They also often generate
noise from their internal digital and analogue workings, which appears as noise
sources between their input terminals and between these terminals and their case or
internal shield. The instrument is designed to be sufficiently immune to this self
generated noise, but the rest of the measurement system connected to the instru
ment may well not be so, particularly if it involves other insufficiently isolated
20 Coaxial electrical circuits for interferencefree measurements
components which provide return paths for interference currents. Figures for this
noise are not usually given in the manufacturers specifications. Possible noise
sources and couplings in a typical instrument are illustrated in Figure 1.28.
The problem is amenable to the general approach given in this section, as
illustrated in Figure 1.29. A mains voltage to mains voltage isolation transformer
must be constructed according to the method given in section 1.1.5, and the
instrument needs to be placed in two nested shields, as shown. In practice, the
conducting case of the instrument will usually be, or can be made to be adequate to
serve as the inner shield. The most significant noise source is often that between the
low input terminal and the instrument case or shield. Its effects can be confined to
noisy
commercial
instrument
measuring
circuit
A
B
mains safety
conductor
input
Hi
Lo
guard
Figure 1.29 Isolating a noisy instrument
DC
power
supply
digital
circuitry
opto
isolators
C
F
10
data
highway
safety
live
neutral
mains:
G
earthor
float
10 k
E
A
B
D
input
Lo
Hi
analogue
input circuitry
guard or
shield
f
Figure 1.28 A typical commercial instrument as a noise source
Introduction 21
the interior of the inner shield by connecting this shield to the low terminal. The
noise source between the low and high input terminals may well be insignif
icant, so that a filter is unnecessary and degradation of the performance of the
instrument can be avoided.
1.1.7 The available methods for isolating outputs
This section addresses the problem of replacing direct connections via conductors
between circuits by indirect coupling methods involving negligible capacitance
between the circuits. Commonconductor coupling, inductive coupling and, with
appropriate shields, capacitive coupling can be eliminated.
Many integratedcircuit packages are available whose internal circuitry is
divided into two separate halves. Information, usually digital, is passed between
them via lightemitting diodes in one circuit and lightsensitive receivers on the
other circuit. The only direct electrical connection between the circuits is via the
high insulation resistance of the substrate and package and the capacitance between
the two circuits. Because of the small dimensions of the circuits, this capacitance is
only of the order of a picofarad. These optoisolators are valuable, though not
perfect, isolation devices. The circuit on the measurement side may produce
unconfined interference of its own. To gain full benefit from the smallness of the
capacitance between the circuits, the integratedcircuit package ought to be arran
ged to be threaded through a hole of minimal size in the measurement screen so that
the plane of the screen coincides with the gap between the two sides of the opto
isolator. Otherwise, interference generated at high potential on the receiving side
may be carried via capacitance to and from the measurement side. Unfortunately,
optoisolators are rarely used with this degree of care.
There are also several commercial examples of these devices where the light
between the two circuits to be isolated from one another is transmitted via optical
fibres. Since the circuits can now be as far apart as desired and the fibres are easily
threaded through small holes in the measurement shield, capacitive coupling can
easily be made to be completely negligible. But in the commercial versions, the
circuit within the measurement screen can create interference as it generates and
receives the light pulses from the other circuit.
This interference has been greatly reduced in an optical fibre data highway
devised by Robinson [2] by two measures.
1. The 20MHz clock needed in the circuitry within the measurement screen to
synchronise the transfer of information with the other circuits on the highway
is totally enclosed within its own nested screen and has no output until data
transfer is requested.
2. The transfer of data on the highway is so arranged and controlled that a given
outstation circuit within a measurement shield is not activated unless the cir
cuitry within that shield is ready to report the results of a measurement.
Therefore, it can often be so arranged that the outstation is totally electrically
quiet, while the circuitry within its measurement shield that it reports on is
actually making measurements.
22 Coaxial electrical circuits for interferencefree measurements
Isolating transformers (see section 1.1.5) have internal screens constructed so
that there is only magnetic coupling between windings. Their purpose is to prevent
any current flowing through capacitance or leakage resistance from one portion of a
network to another.
Electronic operational amplifiers can provide isolation if connection via con
ductors, which carry negligible current, is permissible. Figure 1.30 is a diagram of a
circuit whose input is a symmetrical pair of FET input operational amplifiers
coupled to an output operational amplifier. By choice of the resistance ratios R
2
/R
1
and R
4
/R
3
, the voltage gain of the circuit can have any desired value less than, equal
to or greater than unity. The circuit is differential, that is, the output voltage U
OC
is
the amplified difference voltage U
AB
and contains no component of the common
mode voltages U
AC
and U
BC
. Because the operational amplifiers have FET input
devices, the input currents need to be only a few picoamperes flowing through a
high input impedance of about 1 pF in parallel with 10
13
W slope resistance. In
addition, there will be the DC input bias currents of the devices, which will be
constant and of the order of 1 pA. The sum of these currents will flow back to the
measurement circuit through the current return connection CC
0
, but will usually
be so small that they will not perturb the measurement circuit significantly.
R
1
A
B
C
O
R
2
R
3
R
3
R
4
R
4
R
2
C'
Figure 1.30 An isolation amplifier where one direct connection is permissible
If no direct conducting connection is permissible, isolating amplifiers perform
the same function for analogue information as do optoisolators for digital infor
mation. The circuitry within the measurement shield is based on a conventional
operational amplifier. Its output is converted to a form suitable for reception by the
circuitry outside the measurement shield without any direct electrical connections
by conductors, with one of a variety of techniques including optical coupling,
conversion of the information to a high frequency followed by transformer cou
pling or coupling via a capacitor of only a few picofarads. Similar feedback tech
niques across the isolation can ensure that the analogue voltages or currents in the
measurement circuit transmitter are accurately represented by those generated to
match them in the receiving circuitry. Conversion of the output of the measurement
circuitry to digital form renders this linearisation unnecessary. An optical fibre data
link can enable the isolation to withstand kilovolts. The isolation is compromised
Introduction 23
only by the impedance between transmitter and receiver, and this is typically only a
few picofarads in parallel with a few gigaohms.
Each of the isolation devices described above needs power to be supplied to
the circuit within the measurement shield in a manner that does not significantly
degrade its isolation. Disposable or rechargeable batteries contained entirely within
the measurement shield can do this, but at the expense and inconvenience of
replacing or recharging them when exhausted. The isolated power supply described
in section 1.1.5 was devised to fulfil this need, but the commercially available
solution provided by isolated DCDC converters may be adequate. These con
verters are similar to the isolation amplifiers described above in that DC power
outside the measurement shield is converted to highfrequency AC and conveyed to
the circuit inside the measurement shield by a highfrequency transformer having
physically separated windings. The AC is reconverted to DC power within the
measurement shield and any residual AC is filtered and returned to the measure
ment shield to prevent its passage back into the measurement circuit. The isolation
is incomplete, being compromised by the windingtowinding capacitance and
leakage resistance of the transformer, which might amount to a few picofarads in
parallel with a few gigaohms, but this is often adequate in practice.
1.1.8 Balancing
An alternative approach to current equalisation, which applies to both DC and AC
as well as when the return current conductors cannot be of low impedance, is to
employ symmetrical balanced circuit design. Consider the example illustrated in
Figure 1.31. Information is sent down a symmetrical conductor pair by sources A
and B, which are of equal amplitude and in antiphase. These sources could, for
example, be the secondary windings of a transformer or a balun device. The
information is received by detecting circuits D
1
and D
2
, which are of equal sensi
tivity and input impedance. The outputs of these detectors are subtracted (because
A and B are in antiphase) in circuit D.
Because of the symmetry of the system, any interfering magnetic flux f
ext
,
which threads the loop formed by the conductor that conveys residual imbalance
current back to the sources, causes equal interference voltages to be registered by
D
1
and D
2
. Subtraction of these voltages by Deliminates them. Any capacitive
interference such as the interfering source C can be conveyed to a screen that is
symmetrical with respect to the conductor pair and will then also be balanced out.
1.1.9 Minimising the effects of insufficiently isolated
commercial instruments
Sections 1.1.5 and 1.1.6 are a counsel of perfection. In very particular circum
stances, when systematic errors caused by rectification of noise by measuring
instruments (a common problem) are significant or when very noisesensitive
instrumentation is involved, the construction of the correctly designed isolating
24 Coaxial electrical circuits for interferencefree measurements
systems described in these sections is very worthwhile, if not mandatory. An
example of the compulsory use of these techniques is the automated operation of
measuring systems involving a SQUID [1,3].
But these specialised transformers and supplies are not yet readily available.
By making use of the fundamental principles of section 1.1.1, we attempt in this
section to suggest practical guidelines for getting the best interferencerejecting
performance from the usual commercial equipment. We can only give general
guidelines because the effort instrument designers make for easy interference
rejection varies considerably and is often not apparent from the instrument speci
fications. Each case must be considered separately and is usually an exercise in
damage limitation. Nevertheless, simple general principles applied with a view to
the relative importance of the effects that should be minimised can often achieve
acceptable results in practice.
Typical instruments (all of which, we assume, are mains powered) are
1. power supplies,
2. signal sources (which can be regarded as AC power supplies) and
3. lowlevel measuring instruments such as sensitive digital voltmeters, phase
sensitive detectors and preamplifiers.
A diagrammatic representation of any of these can, for the present purpose, be
drawn as shown in Figure 1.28. The Hi and Lo terminals may very well be the
inner and outer connections made by a coaxial socket.
Some guidelines for obtaining optimum noise rejection are as follows:
1. Eliminate commonconductor couplings in the external wiring. Remember that
data highway cables plugged in later might violate a carefully thoughtout
circuit.
2. Consider whether the circuit connected to the input of a lowlevel measuring
instrument is of high or lowoutput impedance.
3. The mutual inductive coupling, which causes problems in lowimpedance
circuits, is perhaps easier to eliminate than capacitive coupling. Use impedance
transforming devices such as operational amplifiers having a lowoutput
impedance, or impedancematching transformers or isolation amplifiers to
f
ext
D
2
D
1
A
B
D = D
1
D
2
C
Figure 1.31 Balancing a conductorpair circuit
Introduction 25
transform to lowimpedance interconnection networks. Then pay particular
attention to eliminating outer conductor loops (see sections 1.1.1 and 2.1.3).
4. Only the minimum number of connections necessary for the functioning of a
circuit should be made. In particular, the connection of everything in sight to
earth is a council of despair. Instead, draw a complete circuit diagram,
remembering that the circuit that exists in reality may not correspond with the
working circuit diagram in small, but vital, details. For example, there may be
conducting or capacitive routes for currents through ostensibly nonelectrical
apparatus such as optical tables and pressure sensors. Routes of this kind
should be drawn explicitly on the circuit diagram, and their effects considered.
1.1.10 The traditional approach to DC and lowfrequency
circuitry versus the currentbalanced conductorpair
coaxial approach
The techniques used in DC sensing and measurement circuits have traditionally been
viewed as completely different from those employed for AC. This is a serious error.
From the point of view of, at least, interference elimination, there is no difference
whatsoever, and DC practice would greatly benefit from being considered from an
AC viewpoint. In particular, coaxial circuitry could be relevant, and although the
current equalisation technique described in section 3.1.1 is effective only at fre
quencies above a few tens of hertz, this is all that is required to eliminate most
interference. Current balancing in a screened, twistedpair network is even more
appropriate for DC circuits. Great benefit can be obtained by attention to the concept
of isolation and to minimal and correct connection to the mains safety conductors.
There are problems in DC measurements in addition to those encountered in
AC measurements. One problem is that of thermoelectric emfs, and fairly rapid
source polarity reversal is needed to eliminate drifts in them. A second problem,
mostly encountered in highimpedance circuitry, is caused by the phenomenon of
dielectric storage of charge in capacitors including interconductor and inter
component insulation. Dielectric materials can store charge within their volume if
subjected to a unidirectional voltage. The discharge current can persist, exponen
tially decaying away in seconds, minutes or even hours once the voltage source is
removed or reversed. This can create a problem in sensitive DC circuitry. Two
instances are (i) if insulated high and lowpotential leads are in close contact
without an intervening screen and (ii) if an analogue operational amplifier is
configured as an integrator with a capacitor in its feedback loop.
It is often possible to design the circuit so that the discharge current takes a
route that does not affect the measurement, usually by providing proper shielding
or a coaxial approach. The concept of direct impedance is very useful in this regard
(see section 4.1.2).
The magnitude of the phenomenon depends on the particular dielectric. Poly
sulphone and polytetraflourideethelene (PTFE) plastics exhibit a much lower
dielectric storage than the more common polythene insulation.
26 Coaxial electrical circuits for interferencefree measurements
1.1.11 Thermoelectric emfs
The combination of temperature gradients and circuits composed of different metals
produces emfs in the microvolt range. These emfs can cause errors when making
lowlevel DC measurements, and if the temperature gradients are time dependent,
the consequent variation of thermoelectric voltage is a source of lowfrequency
noise.
In principle, the problem can be eliminated from lowlevel DC measurements
by reversing the polarity of the source of the desired signal. If the measuring circuit
is linear, from the superposition theorem, only that part of the circuit response due
to the source will reverse. Taking the mean of the difference between the response
with the source forward and the source reversed then eliminates unwanted stray
voltages. A similar result can be obtained from the difference between the source
forward and source off conditions. If the source cannot be turned off or reversed
rapidly, thermoelectric voltages must be minimised by reducing temperature gra
dients and ensuring that dissimilar metals are eliminated as much as possible from
the circuit. Input terminals to instruments can be lagged and protected from
draughts to reduce temperature differences between them, and coppertocopper
connections used elsewhere in the circuit.
For applications where reduction of thermoelectric emfs to the nanovolt level
is important, it may be necessary to use oxygenfree highpurity copper, or mate
rials with proven low thermal emf with respect to it, everywhere in the circuit. The
unstressed crystalline structure of the conductors must not be compromised by
excessive bending, stretching or work hardening. Permanent or temporary con
nections around the circuit need special attention. Switch contacts used for this kind
of work should be made from special alloys or silver or goldplated pure copper. If
a rubbing action accompanies making a contact, several seconds may be needed for
induced thermoelectric emfs to die away.
Softsoldered joints need careful treatment. Ordinary soft solder as used for
electrical connections has a high thermoelectric emf with respect to copper. It can
nevertheless be used if the copper conductors to be joined are first brought into
intimate contact by applying pressure while being soldered together. If there is no
great excess of solder surrounding the resulting joint, thermoelectric voltages will
be minimal because the soldercopper voltages are shorted by the lowerresistance
coppercopper contact. This approach is, in our experience, preferable to using
lowthermal solder containing cadmium as it is difficult to obtain an electrically
or mechanically sound joint with this material. Pure copper crimped joints are an
alternative to soldering as the stress in the metal is only localised.
Thermoelectric effects in the internal defining point junctions of standard
resistors of value of the order of 1 W or less can be unexpectedly troublesome.
Peltier effects resulting from the flow of measuring currents through them can
cause local heating or cooling. The concomitant thermoelectric emfs generated
appear at the potential terminals so that the apparent value of the resistor can be
altered by several parts in 10
7
for a 1W resistor, an amount that increases in pro
portion to the decrease in value of smaller resistances. The alteration decays away
Introduction 27
in time Dt between the previous application or reversal of the measuring current
and when the measurement is made. Dt depends on the thermal capacity of the
defining point junctions and is typically of the order of a second.
1.1.12 Designing temperaturecontrolled enclosures
In highaccuracy work voltage, current and impedance components, which are as
stable as possible, are a prime necessity. Unstable components hamper investiga
tion by making changes to the type B (systematic) errors of the measurement sys
tem. Changes should produce an anticipated alteration of the result, and obtaining a
different result from that anticipated should indicate the presence of a systematic
error rather than an alteration of the value of an unstable component.
Temperature is usually the main environmental effect that changes the value of
a component. A temperature coefficient producing a relative change of about
10
5
/
(1/Z
1
1/Z
2
)
1
. If Z
1
and Z
2
are resistances R
1
and R
2
, R
out
(1/R
1
1/R
2
)
1
.
The total shunt admittance presented to the input of the detector is Y
total
Y
jwC
D
1/R
D
, where Y is the sum admittance of all cables connected between the
network components comprising Z
out
and the detector input, C
D
is the input capa
citance of the detector and R
D
is its shunt input resistance.
The sensitivity of the detector is reduced by the sensitivity factor s, defined as
the ratio of the voltage actually detected to that detected by an ideal detector of
negligible shunt admittance, connected to the network by cables of negligible shunt
capacitance. Hence, s 1/(1 Y
total
Z
out
), where s is a complex quantity whose
components govern the inphase and quadrature response of the detector. If the
output impedance of the network is a resistance R
out
and the cable and detector
admittances are purely a capacitance C
total
, s 1/(1 R
out
/R
D
jwC
total
R
out
). If
the network is a capacitance ratio bridge as described in section 5.1.2, the corre
sponding expression is s 1/(1 C
total
/C
out
).
2.1.2 The sensitivity of detectors to harmonic content
Bridge networks can be divided into two types for the purpose of analysing the
departure of detectors from the ideal. Some bridge networks have a balance condition
that is independent of frequency, apart from minor departures arising from small
frequency dependencies of components, but others have an inherently frequency
dependent balance condition. The twinT bridge network drawn in Figure 2.2 is an
example of the latter, as its balance condition is wRC 1. The quadrature bridge
described in sections 9.2.19.2.3 is another example.
R R
R/2
C C
2C
D
Figure 2.2 A twinT bridge
If a bridge is nearly frequency independent, there will be only a small amount of
higher harmonics appearing at the detector terminals when the balance condition is
satisfied for the fundamental frequency w
0
. This amount will be proportional to the
product of the harmonic content of the source and the residual frequency depen
dence of the bridge balance. Configuring the detector to detect 2w
0
, 3w
0
, etc. will
indicate the presence of these higher harmonics, but the only consequence of them
is a possible increase in the background indication when the detector is configured
to respond to w
0
. The presence of this background increases the difficulty of jud
ging the minimum indication of a simple tuned detector, but this condition still
D
+
Z
out
U
out
U
in
Y
L
C
D
R
D
preamplifier circuit lead detector
Figure 2.1 The output impedance of a network and the shunt admittance of an
input cable and detector affect sensitivity
34 Coaxial electrical circuits for interferencefree measurements
corresponds to the true bridge balance. The effect is usually negligible if a phase
sensitive detector is used rather than a tuned detector.
A more subtle problem arises if the amplification stages of the detector, before
any frequency selection takes place, are nonlinear. If there are successive harmo
nics present at the input to the detector whose frequencies differ by w
0
, then the
nonlinearity will produce signals at the various sum and difference frequencies,
and in particular, a signal actually at w
0
. An erroneous balance will result. For
tunately, good presentday designs of amplifiers are such that any nonlinearity is
rarely large enough to cause any trouble when used with frequencyindependent
bridges.
Formally, we can see how the fundamental frequency arises from mixing
harmonics if we take the detector response r to contain higher terms than linear in
the applied voltage:
r a
1
U a
2
U
2
a
3
U
3
a
n
U
n
2:1
where the coefficients a
n
usually decrease rapidly with increasing n so that only the
first two terms of (2.1) need be considered.
Then, if the applied voltage contains harmonics,
U
X
1
k0
a
k
sink 1w
0
t 2:2
On substituting (2.2) into (2.1), we find terms like
a
2
a
k
sink 1w
0
ta
k1
sink 2w
0
t
which can be rewritten as
a
2
a
k
a
k1
cosw
0
t cos2k 3w
0
t
2
2:3
The cos[w
0
t] terms represent the spurious signal at the fundamental frequency.
The phenomenon is called intermodulation distortion, and the magnitude of the
effect is proportional to the product of amplitudes of two successive harmonics. A
simple test for its presence is to observe whether there is any change in the bridge
balance when the size of the input signal to the detector is altered with a linear
attenuator. Network components whose values are slightly voltage dependent will
give a similar effect if the network source voltage is changed, and so the test of
deliberately increasing the harmonic content of the source while maintaining its
amplitude unchanged is also valuable.
When the balance condition of a bridge network is frequency dependent, the
network will pass the harmonics present in the source through to the detector
relatively unattenuated, while at balance, the amount of fundamental frequency
should be zero. The problems of insufficient frequency selectivity and nonlinearity
Sources, detectors, cables and connectors 35
of the detector are then far worse, and it is necessary in work of the highest accuracy
to precede the detector with a frequencyselective circuit of sufficient linearity so that
this circuit itself does not cause intermodulation. The magnitudes of harmonics in a
source usually decrease as the harmonic order increases, so that it is often sufficient
to eliminate the second and third harmonics to a high degree of perfection by tuned
filters and to reduce higher harmonics with a passive lowpass filter. Figure 2.3a is
such a filter. The 0.1 H inductors together with the adjustable 100 W resistors and
adjustable capacitors constitute one arm of two nested Wheatstone bridges whose
other arms are the 6800 and 3300 pF capacitors and the 100 W fixedvalue resistor.
When adjusted, these bridges, in principle, reject both phases of the second and third
harmonics completely. Higher harmonics are rejected by the input transformer and
the inductorcapacitor (LC) circuit at the output, both of which are tuned to pass w
0
.
The component values given are suitable for matching the 100 kW output impedance
of a room temperature quadrature bridge to a detector. This filter adds some noise but
it is only of the same order of that inherent in the bridge.
3:1w
0
tuned
in
2w
0
rejection
3w
0
rejection
w
0
out
0
.
1
H
1
0
0
3
3
0
0
p
F
6
8
0
0
p
F
0
.
1
H
1
0
0
1
0
0
0
.
1
H
w tuned
frequency 2w
0
3w
0
(a)
(b)
U
out
/U
in
Figure 2.3 (a) A harmonicrejecting filter. All cores are highQ amorphous metal
material. (b) Its frequency response
The quadrature bridge described in section 9.3.7 in connection with the quantum
Hall effect has components at a cryogenic temperature and a consequent much
lower noise. To take full advantage of this, a different filter is needed in which the
most troublesome harmonics are greatly attenuated by simple seriestuned circuits,
as shown in Figure 2.4. To achieve lownoise performance, the inductors must be
wound on lowloss toroidal amorphous metal cores capable of yielding a Qfactor
of several hundreds, and the capacitors must have a dielectric loss factor of less
than 10
3
. The detector following the filter must reject other harmonics adequately.
36 Coaxial electrical circuits for interferencefree measurements
2.1.3 Noise and noise matching a detector to a network
There is an extensive literature concerned with noise in electrical circuits (see e.g.
Reference 1 or a more recent Reference 2). Here, we discuss only some simple
considerations, which are often sufficient for our purpose.
It is usually easy to amplify a small output signal from a network sufficiently
so that the sole criterion to be investigated is the optimisation of the signaltonoise
ratio of the amplifier. When we are concerned in this book with networks con
taining only passive components, we need consider only Johnson noise. (We will
assume that troublesome, but in principle avoidable, noise from interference,
microphony, etc. has been eliminated.) Johnson noise generates a power
U
2
n
R
i
2
n
R 4kTDf 2:4
in a resistance R. k is Boltzmann constant (&1.38 10
23
J/K), T is the absolute
temperature (&300 K for room temperature) and Df (Hz) is the bandwidth of the
detecting device, so the noise power is about 1.6 10
23
W for a 1Hz bandwidth
at room temperature. For a network connected to a detector, the resistance R con
cerned represents the total shunt loss; that is, all sources of energy loss in the
network as well as in the detector are included in it. The noise performance of most
AC detectors now approaches the Johnson noise limit over a range of source
impedances presented to their inputs. Optimising their performance is therefore a
matter of impedance matching of the network output impedance to an impedance
that is within the optimum input impedance of the detector.
The noise situation can be analysed by replacing the actual resistors and other
loss sources with equivalent noise sources in series with ideal noisefree resistors
when we are considering a voltage noise source or in parallel when we are con
sidering a current noise source. The superposition theorem then allows us to con
sider the effect on a detector of the signal source and the various noise sources
separately.
Noise arising from components that have an outer coating of lossy dielectric,
which forms a lossy capacitance between the inner and outer conductors, can be
eliminated by surrounding the components with an intermediate conducting casing
supported by lowloss dielectric, which is connected to the inner conductor, as shown
in Figure 2.5.
2w
0
reject 3w
0
reject 4w
0
reject
Figure 2.4 An ultralow noise harmonicrejecting filter for use with a bridge
having components at cryogenic temperatures
Sources, detectors, cables and connectors 37
(b)
high low
(a)
high low
Figure 2.5 (a) A component coated with a lossy dielectric. (b) Removing the loss
contribution to the noise
In this section and the following section, we assume that all impedances are
pure resistances in order to bring out the point we wish to make as simply as
possible. We also take as a simple model of a real detector the equivalent circuit of
Figure 2.6 where D is an ideal noisefree detector of high input impedance, that is,
of negligible input admittance. Its actual noise resistance and input impedance are
represented by a shunt conductance G
D
. We suppose that the total loss of the net
work can be represented by a single conductance G
S
, as would be the case for a
network having only resistive components and negligible shunt admittance.
G
D
D
G
S
U
S
I
S
Figure 2.6 The equivalent circuit of a detector connected to a noise source
The total noise admittance presented to D is G G
S
G
D
. The noise power is
U
2
n
G 4kTDf 2:5
The signal current in the loop is
I
S
U
S
G
S
G
D
G
S
G
D
U
S
G
S
G
D
G
2:6
The voltage across the detector is
U
D
I
S
G
D
U
S
G
S
G
2:7
Hence the signaltonoise power ratio
S
U
2
D
U
2
n
U
2
S
G
S
G
S
=G
4kTDf
2:8
38 Coaxial electrical circuits for interferencefree measurements
Another source of noise in addition to Johnson noise occurs at low frequencies
in nonmetallic conductors and semiconductor devices. This is termed flicker
noise or 1/f noise because its magnitude is approximately inversely proportional
to the frequency f for unit bandwidth; it is unlikely to be detectable at 1 kHz but
will increase in relative importance as the frequency is reduced.
2.1.4 The concept of a noise figure
P U
S
2
G
S
/4 is the maximum possible signal power that can be transmitted into the
detector terminals by varying G
D
until it equals G
S
. The signaltonoise ratio
obtained when the value of P equals the total noise power 4kTDf is called the noise
figure or factor n; it is a measure of by how much the system fails to meet the ideal
signaltonoise ratio. In the above instance,
n
G
G
S
G
S
G
D
G
S
2:9
Often N 10 log
10
n is used instead of n and the condition of optimum power
transfer into the detector (G
D
G
S
) gives n 2 or N 3.01 dB.
In the present example, it is instructive to note that
n !1 or N !0 dB as G
D
!0 or G
S
!1 2:10
that is, the best signaltonoise ratio is not obtained under conditions of maximum
signal power transfer. Thus, in general, it is important not to confuse power
matching with noise matching; it is noise matching that is usually relevant.
For a real detector, G
D
is often fixed by the instrument design, but by interposing
a matching network between the source and the detector, provided that it itself does
not introduce further significant noise, the apparent value of G
S
presented to the
detector terminals can be altered. Two such matching devices that can be used either
separately or in combination are a tuned LC circuit and a ratio transformer.
In Figure 2.7a, we illustrate a typical problem of matching a network modelled
as a source having an output impedance Z
n
, the real component of which generates
noise, to a preamplifier having both a noise voltage U
N
and a noise current I
N
.
A narrowband tuned LC circuit (Figure 2.7b) can simultaneously perform the
useful task of rejecting harmonic components of U
S.
A ratio autotransformer
(Figure 2.7c) is a broadband simple matching device from low to high impedance
as shown, or, by reversing input and output connections, from high to low impe
dance (Figure 2.7d). A ratio transformer (Figure 2.7e; section 1.1.5) could also
provide isolation of the detector from the network. Harmonic rejection can be
accomplished by tuning the transformer with a capacitor that shunts either the
primary or secondary winding. In a practical transformer, it is the inductance of
each winding that makes tuning possible. The inductor or transformer core should
have a low loss so as not to introduce further noise.
From the discussion so far, it might be thought that using a matching device to
make the network impedance presented to the detector look very low would be
Sources, detectors, cables and connectors 39
sufficient to resolve all signaltonoise problems. Unfortunately, this is not so
because we have adopted too simple model for a real detector; its noise attributes
can rarely be represented by a simple conductance G
D
.
Figure 2.8 is a better representation of a real detector. An ideal detector
(noiseless and of infinite input impedance) is shunted by a noiseless admittance
Y
D
in series with a voltage noise U
D
, the whole being shunted by a current noise
source I
D
.
resonant circuit
matching transformer
(b)
(c)
C
L
(e)
network
network
preamplifier
preamplifier
isolating transformer network preamplifier
matching transformer
(d)
network preamplifier
+

D
generator
generator
generator
generator
generator
preamplifier
U
G
U
n
U
N
I
N
+

D
U
N
I
N
+

D
detector
detector
detector
detector
detector
U
N
I
N
+

D
U
N
I
N
+

D
U
N
I
N
Z
n
U
n
Z
n
U
n
Z
n
U
n
Z
n
U
n
Z
n
U
G
U
G
U
G
U
G
N
2
N
2
N
2
N
1
N
1
N
1
network
(a)
Figure 2.7 Circuits to match a detector to a network for better sensitivity:
(a) a source, the equivalent circuit of a network, the equivalent noise
representation of a preamplifier and a final detector; (b) and (c)
matching a lowimpedance source and network to a highimpedance
detector; (d) matching a highimpedance source and network to a low
impedance detector; (e) an isolating matching transformer
40 Coaxial electrical circuits for interferencefree measurements
D
I
D
Y
D
U
D
Figure 2.8 The equivalent circuit of a detector
If, therefore, a detector for which this representation is valid is connected to a
network by a matching circuit of low output impedance and low signal voltage but
high signal current so that I
D
and Y
D
are shorted out, the detector voltage noise U
D
is still registered and can swamp a small signal voltage. If the matching circuit is
designed to have high output impedance, high signal voltage and low signal cur
rent, the effect of U
D
may well be negligible but the signal current will be swamped
by I
D
in parallel with it. There is an optimum between these two extremes.
U
D
and I
D
are easily measured for a given detector. Having calibrated the
sensitivity of the detector for both voltage and current by connecting it to standard
signal generators and attenuators, U
D
can be measured as the voltage indicated
when the input terminals are shortcircuited. I
D
is the voltage indicated times the
input admittance of the detector Y
D
when the input terminals are opencircuited and
screened to prevent electric field pickup.
The optimum signaltonoise situation arises when the voltage generated by I
D
flowing in the source impedance Y
S
added to Y
D
equals the voltage noise U
D
.
Manufacturers of detectors often supply noise contours, that is, loci of con
stant noise figures on a graph of source frequency against source impedance. From
these, the optimum impedance to be presented to the detector at a given frequency
can be deduced and a suitable matching network designed.
2.2 Attributes of sources
A network source will usually consist of a signal generator, possibly followed by a
power amplifier and a power transformer if the output power of the generator is
insufficient. It is useful to combine the functions of matching and isolation (by
means of properly constructed screens see section 1.1.5) in one device. The
arrangement must be capable of delivering the necessary power without excessive
distortion, but if the output goes straight into a transformer forming part of a
measurement network, there is no point in striving to achieve excessively low
distortion because the transformer will add some distortion on its own account. This
can be minimised for a voltage transformer if the output impedance of the source is
made very low compared to the input impedance of the transformer. A transformer
adds distortion because the instantaneous input impedance of a transformer
depends on the relative permeability of the core, which in turn depends on the flux
level and therefore on the excitation level. This variation of its instantaneous input
Sources, detectors, cables and connectors 41
impedance causes a varying voltage drop across the output impedance of the
source, and this distorts the waveform. Attaining a total harmonic distortion of
0.1% or less is reasonable.
If a bridge balance is frequency dependent, the source must have the necessary
frequency stability. This is easily attained if the source is a frequency synthesiser
locked to a precision frequency standard. The frequency can be measured by an
electronic counter also locked to a precision frequency. For lowfrequency mea
surement, better resolution is obtained by operating the counter in the period mode.
2.3 Properties of different detectors
Phasesensitive detectors (also called lockin detectors or vector voltmeters) are
most commonly employed for observing small signals and for balancing bridge
networks but wideband and simple tuned detectors are also useful, particularly for
diagnostic purposes. To obtain an optimum signaltonoise ratio an appropriate
preamplifier is needed.
2.3.1 Preamplifiers
Preamplifiers are needed for all kinds of electrical measuring circuits. They can be
isolated by being powered by batteries contained within the conducting case of the
preamplifier or have a mainspowered supply. A mainspowered supply must be
properly isolated (see section 1.1.5) to eliminate any net current into the input of
the preamplifier.
The preamplifier shown in the circuit of Figure 2.9 is especially useful for high
source impedances, but it also performs well for other values of source impedance.
The input transistor is an njunction fieldeffect transistor (JFET), for example, type
2SK170 made by Toshiba. The input resistor R
in
is of the order of a gigaohm, so,
being in parallel with the source impedance, it does not usually significantly
increase the noise. Also, the noise current of the transistor must be small enough
not to generate significant noise when it flows through the source impedance.
U
S
R
D1
R
in
R
a R
F
R
S
R
V
C
in
R
D2
in
out
+

+

D
T
Figure 2.9 A lownoise preamplifier suitable for highimpedance sources
42 Coaxial electrical circuits for interferencefree measurements
Typical values for the equivalent input noise voltage and noise current are 1.5 nV/HHz
and a few fA/HHz, respectively. Preamplifiers for optimum noise matching to high
output impedances need special attention to avoid including lossy dielectric
materials in all capacitances to the screen in the region of their input circuitry. In this
example, this applies to all materials within the shaded area. A twostage feedback
amplifier sets a stable gain as the ratio of the feedback resistor R
F
and the transistor
source resistor R
S
. A gain of 100 is typical.
Figure 2.10 is a lownoise preamplifier circuit [3] optimised for cryogenic
source impedances of the order of 10 kW such as that of the quadrature bridge
involving two cryogenic Hall resistances described in section 9.3.7. This bridge
network has an output impedance of 12.9 kW (at a temperature of the order of 1 K)
in parallel with a virtually noisefree capacitance of 10 nF so that the noise
amplitude is of the order of 1 nV/HHz. Using a matching transformer (see section
2.1.3) in conjunction with a lownoise highimpedance preamplifier is not prac
ticable because of the excess noise generated by transformer core losses.
The preamplifier achieves low voltage noise by using multiple parallel
connected JFETs having a common drain resistor and individual source resistors.
The bootstrap action of the amplifier increases the drain resistance virtually by
means of the feedback capacitor C
F
. Optimum performance is obtained by selection
of the individual JFETs for low noise and, for example, four parallel JFETs can
produce a noise voltage of only 0.5 nV/HHz.
2.3.2 Wideband (untuned) detectors
The most useful example of these is the oscilloscope, which can reveal the presence
of unwanted interference. A newly constructed network should always be scruti
nised in this way, and it is prudent to reexamine it at intervals also.
in
out
T
1
T
N
x1
multiple parallel connected JFETs
main
amp
bootstrap
amp
U
S
R
D1
R
S1
R
in
R
F
R
SN
C
in
C
1
C
N
C
F
R
N
R
D2
Figure 2.10 A lownoise preamplifier suitable for source impedances of the order
of 10 kW
Sources, detectors, cables and connectors 43
2.3.3 Narrowband (tuned) detectors
These often incorporate a frequencyselective network (usually a twinT bridge,
Figure 2.2) in the feedback loop of an electronic amplifier. The resulting frequency
response is similar to that of an LC tuned circuit; that is, it attenuates at fre
quencies away from the resonant frequency by a rate governed by an effective Q of
the circuit. Successive harmonics are attenuated with respect to the preceding one
by a factor of the order of Q. It is difficult to design stable circuits with Qvalues
much greater than 10
2
10
3
, so the detector will have for the second harmonic 2w
0
not much less than 1% of the sensitivity that it has for the fundamental frequency
w
0
. As we have noted in section 2.1.2, this situation can cause errors in detecting a
bridge balance condition.
Further, a simple tuned detector responds to both inphase and quadrature
signals alike; the indication corresponds to the square root of the sumofsquares of
the inphase and quadrature components. This can make the balancing of a bridge
network where the supposed inphase and quadrature adjustments are not in fact
orthogonal somewhat difficult. A degree of phase dependence can be obtained by
deliberately introducing a signal of the desired phase directly into the detector from
the network source. Smaller quadrature signals will then produce very little addi
tional indication, but inphase signals will be fully displayed as an addition to the
standing deflection.
A simple extension of this technique can improve the sensitivity in balancing a
bridge network to almost the limit imposed by detector noise by taking as the
balance point the mean of two adjustments, which produce equal relatively large
indications on either side of the correct balance condition. This technique is
superior to simply trying to adjust for a minimum indication because it gives the
detector a modicum of phase sensitivity against both a small quadrature component
and noise.
2.3.4 Phasesensitive detectors that employ a switching technique
All phasesensitive detectors need a reference signal; in bridge networks it is
usually derived from the bridge source. In the past, many phasesensitive detectors
applied the reference signal to switching circuitry to rectify the input signal. Since
they are phase related, any harmonics present in the input signal will also be rec
tified. The rectification of even harmonics sums to zero but the last of the n cycles
of the nth harmonic in a rectification cycle does not cancel so that the odd nth
harmonic is detected with an efficiency of 1/n relative to that of the fundamental
frequency. Therefore, switching phasesensitive detectors need to be used with
caution, particularly in bridge networks. If the bridge balance condition is fre
quency dependent (see section 2.1.2), they must be preceded by a tuned filter to
reject harmonics sufficiently.
2.3.5 Phasesensitive detectors employing a modulating technique
Newer phasesensitive detectors in effect multiply the incoming signal and the
reference signal, and display the result after passage through a lowpass filter. The
44 Coaxial electrical circuits for interferencefree measurements
effect on all signals other than that at the reference frequency is a timevarying
output at the difference frequency, and this is rejected by the lowpass filter.
Therefore, there is, in principle, nearly complete rejection of all harmonics except
the one desired. Naturally, the reference signal must not contain any significant
harmonic content. Also, nonlinearity at small signal levels and internal harmonic
generation as saturation levels are approached lead to some likelihood of sig
nificant intermodulation distortion, so checks for all these effects should be made.
The design philosophy of most signaldetecting instruments now is to convert
an analogue input to digital form as soon as is practicable after some amplification
at the input stage. Any signal recovery, manipulation, filtering, etc., including
multiplication with a similarly digitised reference signal, is then carried out by
mathematical operations on the digital representation of the signal. This approach
has the potential problem of aliasing [4] if the variation rate of the signal approa
ches the digitisation rate, but it results in better linearity.
2.4 Cables and connectors
The properties of coaxial cable used for highaccuracy work need careful con
sideration. Where networks containing small direct admittances (such as small
valued capacitors) are concerned, any loss in the cable dielectric, particularly of
cables making connections to the detector and any associated subnetwork, will add
noise. Polythene and PTFE are good lowloss dielectrics. If the capacitance per unit
length of the cable is unduly large, the available outofbalance signal will be
shunted and the balance sensitivity will be decreased (see section 2.1.1).
In some cables, the central conductor is of steel (to increase its mechanical
strength) plated with copper. This increases its series resistance but has no other
consequence.
Coaxial cables suffer from microphony at audio frequencies, and, in an extreme
case, acoustic oscillations in a cable or a component carrying a relatively large
current or voltage can be transmitted to a touching cable associated with a detector
where they can be reconverted into an electrical signal. Therefore, it is better if all
cables are of antimicrophonic construction. This is usually accomplished by putting
a layer of conducting plastic between the dielectric and the outer conductor where
microphonic signals are generated or reconverted by the triboelectric effect. Such a
layer also prevents leakage of electric field through holes in the woven outer con
ductor. In some examples, the resistivity of the plastic layer can be unduly high and
will thus cause some loss and constitute another source of noise, particularly at higher
frequencies.
Therefore, the choice of cable at audio frequencies involves a compromise
between low shunt capacitance, low series inductance, low series resistance, reason
able physical size, flexibility and cost. The authors have made extensive use of PTFE
dielectric cable of about 6 mm overall diameter that has a characteristic impedance of
75 W, a capacitance of 70 pF/m having a loss angle of 35 10
6
radians and a series
inductance and resistance of 0.4 mH/m and 0.07 W/m, respectively. Microphonic
Sources, detectors, cables and connectors 45
effects are reduced by a coating of conducting varnish applied to the dielectric. This
varnish is easily removed locally when terminating the cable with a connector.
The principal requirements for coaxial connectors are that the series resistance
they introduce between their mating surfaces should be low and reproducible, and
they should have low loss in any dielectric surrounding the inner conductor. The
inner conductor should be completely screened by the surrounding outer conductor
when the connectors are mated, and there should be available Tconnectors of low
impedance. Laboratories undertaking work of the highest accuracy at audio fre
quencies usually use the British Post Office (BPO) MUSA connectors that, besides
the above desirable properties, are simple to attach to cable ends and components,
are relatively inexpensive, and mate and unmate with a simple push or pull. The
more readily available Bayonet NeillConcelman (BNC) connectors suffer from a
higher and more variable contact resistance. Ordinary readymade cables with BNC
connectors do not usually have an antimicrophonic conducting layer. General
Radio GR900 connectors are unnecessarily elaborate and expensive for lower
frequency applications. The hermaphrodite GR874 connectors are suitable but are
somewhat bulky. Care must be taken to ensure that the central mating conductors
do not become partially unscrewed and provide a source of unwelcome variable
resistance. At higher frequencies, the subminiature SMA, B and C connector series
and readymade cables are good and convenient because they are impedance
matched to a cable. Care must be taken when mating and unmating them to provide
only axial force or else the life of the connectors will be very limited.
References
1. Robinson F.N.H. Noise and fluctuations in electronic devices and circuits.
Oxford: Clarendon; 1974
2. Ramm G., Bachmair H. Optimisation of the signaltonoise ratio of AC
bridges. Tech. Mess. 1982;49:32124
3. Schurr J., Moser H., Pierz K., Ramm G., Kibble B.P. JohnsonNyquist noise of
the quantised Hall resistance. To be published in IEEE Trans. Instrum. Meas.
2011
4. A Guide to Measuring Direct and Alternating Current and Voltage Below
1 MHz. London: The Institute of Measurement and Control; 2003
46 Coaxial electrical circuits for interferencefree measurements
Chapter 3
The concept of a lowfrequency coaxial network
Interferencefree balancedcurrent circuitry evolved from the need to develop bridge
measurements of standard impedances to obtain ever higher accuracies to satisfy
demands for better characterisation of passive electronic and electrical circuit
components.
Early comparisons of the value of impedances were confined to resistance
measurement because only direct current sources and detectors were available. A
little later, some form of commutator provided intermittent or reversed current, and
this allowed capacitances and mutual and selfinductances (which are only defined
for the passage of timevarying currents) to be compared. Bridge design continued
to evolve around DC techniques for which the spatial disposition of conductors,
impedances, detectors and sources is immaterial (interference considerations apart).
The basic standard of impedance was a DC resistor, assumed or designed to have
nearly the same value with AC.
Even after sinusoidal sources and detectors responding to a single frequency
became available and some of the advantages of AC bridge comparisons became
apparent, the past history of the subject lingered on in the form of singleconductor
networks, but the advent of the calculable mutual inductor (see section 6.4.12) and
the even more accurate calculable capacitor (see section 6.2) necessitated the
development of better techniques.
Electrical impedances store or dissipate the energy of electric and magnetic
fields and currents, and so for a reproducible, welldefined standard of impedance,
the location and strength of its fields must be defined by fixing the boundary of the
space they occupy. A bridge network relates the energy flow through two or more
sets of impedance standards in order to compare their values accurately; to do this,
the standards need to be interconnected by conductors to form a network. Therefore,
the electric and magnetic fields surrounding these conductors must also be controlled
and accounted for. If the impedance standards have exposed terminals and single
wire conductors interconnect them, fields surround both conductors and standards.
The strength and distribution of these fields are altered by changes in the relative
position of the standards, conductors and external objects, including the experi
menter. The use of such networks inevitably leads to imprecise relationships between
the values of the standards because the stray fields associated with the standards and
networks are altered by changes in their environment.
Therefore, the subject of accurate impedance measurement was developed by
using coaxial cables which carry equal and opposite currents in their inner and
concentric outer shielding conductors. Components and standards were encased in
electric and magnetic screens. Fields are thereby confined to the inside of cables
and standards, and a bridge network becomes well defined. Meaningful measure
ments can then be made with relative uncertainties of as little as 1 in 10
9
,
if
required.
This approach may be viewed simply as the application of the coaxial tech
niques common in radiofrequency work to lowfrequency networks, combined
with more careful control of the fields associated with circuit components. The
ideas have been developed by many people over many years, but principally by
R.D. Cutkosky, at the National Institute of Standards and Technology, USA, and
A.M. Thompson, at the National Measurement Laboratory, Australia.
One way of looking at coaxial networks is to see them as two superimposed
networks. The first network consists of the meshes of components and the con
ductors connecting them. The second network comprises the screens of the com
ponents and the outer coaxial screen of the interconnecting cables. The
configurations of the two networks are identical, and, by providing each mesh with
an equalising device such as a current equaliser (see section 3.1.1) or a differential
input amplifier (section 1.1.7), the current in the outer screen of each cable and
component is constrained to be equal and opposite to the current in the components
and central conductor. The current in any cable as a whole is then zero over the
frequency range of equalisation and no external magnetic field is created. The
second network of screens and cable outer conductors has only a small impedance
and consequently is all at nearly the same potential, so that there is no significant
external electric field either. Consequently, the layout of cables and components
has no effect on the performance of the circuits. Also, the reciprocity theorem states
that such networks also do not respond to fields from external sources; that is, they
are interference free. Because of these desirable properties, the purpose of this book
is to encourage coaxial techniques for DC and AC circuitry up to at least 1 MHz.
Unexpected interactions between apparently unrelated parts of a circuit are a fre
quent source of frustration, if not outright design failure. They can be avoided by
visualising and minimising the electric and magnetic fields associated with the
various parts of the circuit. The role of power supply leads and their associated
current return conductors (earth or ground conductors) to and from parts of
circuits needs careful consideration in relation to the network as a whole.
The tendency now in designing electronic instruments is to convert the output
of very simple analogue input circuitry to digital form by periodic sampling of its
amplitude as early as possible in the chain from input to final presentation of the
result. Any manipulation of the signal, for example, for bandwidth or bandpass
filtering or for combining with another input for phasesensitive detection is carried
out by arithmetic algorithms applied to the numbers representing the signal. This
approach has frequency limitations and other difficulties because the bandwidth is
inherently limited to half the sampling frequency. At greater frequencies, false
48 Coaxial electrical circuits for interferencefree measurements
results are produced (see Reference 1). These considerations are outside the scope
of this book, which is concerned only with the correct gathering of the input signal,
and, in particular, in avoiding its corruption by noise, interference and crosstalk
from other input channels. It is true that some of these effects can be minimised in
the signalprocessing chain by filtering, etc., but it is a much better and cost
effective practice to eliminate these problems from DC to 100s of MHz DC and AC
circuitry in the first place (see Reference 2).
3.1 The coaxial conductor
In section 1.1.1, we proved that for an infinitely long straight coaxial cable in
which, by some means, the current through the outer conductor is maintained at all
times equal and opposite to the current in the inner conductor, there is no external
magnetic field at all. In practice, cables are neither infinitely long nor straight or
perfectly coaxial, but the external magnetic field of a finite and curved length is
still usually negligible, provided the length and radius of curvature are large com
pared with the diameter of the cable conditions usually fulfilled by virtue of its
mechanical stiffness.
In a network of components interconnected by balancedcurrent coaxial cables,
the outer conductors should be electrically continuous and without any appreciable
impedances in their network. Therefore, despite the currents flowing in them, they
form a nearly equipotential surface and the electric field outside the cables is also
negligibly small. Therefore, we have a system of conductors outside which there is
no appreciable electromagnetic field. That is, the network of cables cannot radiate
interference, and according to the reciprocity theorem, it cannot respond to external
interference either, and we have an interferenceimmune system.
It cannot be too strongly reemphasised that there is nothing different about
similarly constructed DC networks in this regard; constructed according to these
principles, they will also be interferenceimmune.
Equality of the opposing currents in the inner and outer conductors of a coaxial
cable in a mesh, where the outer conductor has low impedance, occurs auto
matically at frequencies higher than a MHz or so, because the goandreturn
impedance comprising the resistance and inductance of the coaxial cable is much
lower than that of the impedance of the loop traversed by the outer conductor.
At lower frequencies, the current equaliser technique described in section 3.1.1 is
necessary. This technique does not equalise direct currents or currents whose fre
quencies are below 50 Hz or so, but it suffices for eliminating interference which
usually is of higher frequency.
3.1.1 Achieving current equalisation
When a single coaxial cable goes to an isolated component, which has no other
external electrical connection to it, the current flowing in the inner conductor of the
cable must return via the outer conductor, and current equalisation is automatic.
The concept of a lowfrequency coaxial network 49
But if the component has other connecting conductors so that it is a part of a mesh
in a network, positive steps must be taken to achieve current equalisation.
A currentequalising technique was developed for kHz frequencies in
conjunction with the coaxial AC bridges described later in this book because it
ensures that the electrical definition of the standards measured by these bridges is
complete. This technique also eliminates from circuit networks in general inter
ference induced by external magnetic fluxes within the working frequency range of
the technique.
Consider, as illustrated in Figure 3.1a, just one mesh of a network of conductor
pairs (usually coaxial cables). The circuit of the inner conductor usually involves a
voltage source in series with a comparatively high impedance Z that determines the
value of the current I
P
flowing around it. The lowimpedance circuit of the outer
shield conductors has no additional impedances or voltage sources. If the cable is
threaded through a highpermeability magnetic transformer core as shown, the core
and mesh become a current transformer. The inner conductor is its highimpedance
primary winding and the outer conductor is the secondary winding, which is
approximately a shortcircuit across which very little voltage can be induced. The
circuit is redrawn in Figure 3.1b to illustrate this. Because there is very little
voltage across the secondary winding, there can be very little flux in the core, and
this can only be so if there is an almost equal and opposite current I
S
in the sec
ondary (i.e. the circuit of the outer conductors) to I
P
in the primary windings
(i.e. the circuit of the inner conductors). The total current in the cable as a whole,
which is the sum of the current in the inner and the current in the outer, is therefore
approximately zero, as required. The effectiveness of the device is greatly
enhanced if the cable is threaded through the core a few times rather than just once,
and this device can achieve an approximate equalisedcurrent condition for
frequencies from about 10 Hz to 1 MHz.
Let f
P
and f
S
be the magnetic fluxes in the core caused by the primary and
secondary currents I
P
and I
S
, respectively, which thread the core n times. The total
flux in the core is
f f
P
f
S
n I
P
I
S
(a) (b)
Z
I
P
I
S
U
z
n
f
n
Z
I
P
I
S
Figure 3.1 (a) One mesh of a coaxial network threaded through a core and (b) the
equivalent circuit
50 Coaxial electrical circuits for interferencefree measurements
Then
n
df
dt
U I
P
Z I
S
z n
2
dI
P
dt
dI
S
dt
jwLI
P
I
S
for a sinusoidal current. L is the inductance of the wound magnetic core, Z is the
impedance of the circuit of inner conductors and z r jwL is the impedance of
the circuit of outer conductors.
Hence,
I
P
I
S
I
S
z
jwL
and
U I
P
Z I
S
z I
P
Z
z
1 z=jwL
% I
P
Z z
z
2
jwL
If z (  jwL
U % I
P
Z z 3:1
and
I
P
% I
S
3:2
Equation (3.1) shows that the apparent impedance of the primary circuit has
been increased by z, whilst (3.2) shows that equality of currents has been achieved
to a good approximation.
The above analysis can be refined by including the energy loss in the core. This
loss can be represented by a resistance R in series with the inductive impedance wL
of the wound core. That is, its Qfactor is equal to wL/R. Typical values for wL and
R are 100 and 10 W, respectively. The fractional current inequality is
I
P
I
S
I
P
z
jwL R
3:3
The apparent impedance of the primary circuit has been increased by
Z z
z
2
jwL R
3:4
If the circuit consists of several meshes, each independent mesh ought to be
provided with an equaliser, but it is irrelevant from the point of view of equalisation
at which point around the mesh it is placed.
The concept of a lowfrequency coaxial network 51
For frequencies in the range above 100 kHz, the resistance of the meshes of
outer conductors becomes small compared with the impedance of the inductance of
their open loops, and this latter impedance is small compared to the impedance of a
goandreturn path through inner and outer conductors of cables. Then current
equalisation occurs without the need for a transformer core.
So to achieve successful equalisation, two conditions must be met:
1. The circuit of one of the two conductors (the outer of coaxial cable conductors
or one of a twisted pair of conductors) must have an impedance that is mainly
its resistance, which is low compared with the inductive impedance of the
wound equaliser.
2. The circuits of the two conductors must nearly coincide in space, that is, the
open area between them must be small.
If a set of equalisers are a component of a balanced coaxial bridge network,
their imperfect operation caused by the finite impedance of the meshes of the outer
conductors means that the currents in the inner and outer conductors of the cables
are not perfectly equal and opposite. This has a small effect on the balance con
dition of bridge networks for the following reasons:
(i) The cable corrections (see section 5.4) assume perfect current equalisation,
and if equalisation is imperfect a fraction of the openloop inductance con
tributes to the series inductance of the cables.
(ii) The contribution of the outer impedance of terminalpair components is
altered. For fourterminalpair components, the relevant impedance is the
fourterminal impedance of the outer conductor that is usually very small so
that the alteration of the fourterminalpair impedance is usually negligible.
(iii) The output voltages between inner and outer conductors of the coaxial
connectors of transformer taps are slightly changed.
(iv) There are residual magnetic couplings between meshes of the network.
A single total correction for all these small effects can be obtained from the
simple procedure of equaliser evaluation described in section 3.2.2. Equaliser eva
luation generates a small correction that can be applied to the measurements to give
results that would be obtained with perfect current equalisation everywhere. Note that
the effect of residual magnetic couplings implies that the relative orientations and
positions of equalisers and meshes should not be altered once evaluation has been
carried out. The need for attention to this point can easily be tested by deliberately
temporarily altering positions and orientations of cables and equalisers to observe
whether the bridge balance is significantly affected.
As noted earlier, the interference rejection properties of a currentequalised
coaxial network follow from the fact that it generates no significant external
magnetic flux. The reciprocity theorem then ensures that the network does not
respond to interference from externally generated fluxes.
A current equaliser is represented symbolically by a toroidal core surrounding
a coaxial cable, as shown in Figure 3.2. The magnetic transformer cores used for
this purpose are usually, but not necessarily, toroidal.
52 Coaxial electrical circuits for interferencefree measurements
(a) (b)
Figure 3.2 (a) A highpermeability toroidal transformer core used as a current
equaliser. (b) Its symbolic representation
3.1.2 The concept of a coaxial network
In this book, we are concerned with conductorpair networks where one network of
the pair has a low impedance compared to the other, and where the current in any
one conductor is matched by an equal and opposite current in the other conductor of
the pair. The potential differences between corresponding nodes of the pair of
networks are quantities of interest because, for example, detectors or sources are to
be connected between a pair of corresponding nodes.
This concept can be generalised to any currentequalised network. With reference
to Figure 3.3, which shows one mesh of an equalisedcurrent conductorpair network,
where the potentials at its nodes, U
A
, U
B
, etc., are referenced to some arbitrary zero,
U
A
U
A
0 U
B
U
B
0 U
A
U
B
U
A
0 U
B
0 % IZ z
The small inequality caused by imperfect equaliser action, denoted by e in the
figure, can be removed by the procedure of equaliser evaluation described in
section 3.2.2.
Usually, the impedances like Z, which represent the impedances of compo
nents in the circuit, have much greater values than those like z, which represent the
impedances of the current return conductors.
Z
z
I
I' = I (1 e)
A
B
B'
C
C'
D
D'
A'
Figure 3.3 An equalisedcurrent conductorpair mesh
The concept of a lowfrequency coaxial network 53
If we replace the small impedances z of the primed mesh with ones of zero
impedance, U
A
0
U
B
0
, and if we then enhance every impedance between two
nodes of the unprimed mesh by the corresponding impedance of the old primed
mesh, an equivalent single mesh with the same nodal equations
U
A
U
B
IZ z
0
; etc: 3:5
as the conductorpair mesh is obtained.
If the conductorpair mesh is, in fact, coaxial circuitry as drawn in Figure 3.4,
the entire primed mesh constitutes an equipotential surface and there are no
external electric fields. Hence, the behaviour of a conductorpair net only differs
from that of the corresponding singleconductor net in the vital respect that a
coaxial conductorpair net has no interactions between its constituent meshes or
between itself and external electromagnetic fields.
B
D
C Z
z
A
I
P
I
S
Figure 3.4 A coaxial version of Figure 3.3
At frequencies of the order of, or greater than, 1 MHz, current equalisation
occurs without the need for equalisers because the openloop inductance of a mesh
is much greater than the goandreturn inductance of a conductor pair.
3.2 Construction and properties of coaxial networks
3.2.1 Equalisers in bridge or other measuring networks
It must be understood that the set of current equalisers in a network is at least as
important as any of the other components (such as ratio transformers, impedances
to be compared, combining networks, etc.). The equalisers need very careful and
detailed attention if the highest possible accuracy and degree of interference
54 Coaxial electrical circuits for interferencefree measurements
elimination are to be achieved. Note that imbalanced mesh currents result from
imperfectly efficient equalisers and affect the main outcome of the measurement
network (see section 3.1.1). The process of assessment of all the equalisers acting
together will quantify this effect (detailed in section 3.2.2). Simply measuring the
unbalanced mesh currents will not enable the total effect of equalisers to be assessed.
There are at present three ways in which equalisers can be constructed:
(i) The simplest and most familiar way is to wind 1020 turns of coaxial cable
through a highpermeability toroidal core (see section 3.1.1 and Figure 3.2).
The equality of the opposing currents in the inner and outer conductors can be
improved by increasing the number of turns, n, on the core because the
inductance of the winding, L, increases as n
2
, whereas the resistive component
of z increases only as n. Enhanced efficiency can therefore be obtained by
winding 2040 or more turns until the window of the toroid is full. Threading
a greater number of turns through the usual toroidal core may require a thinner
coaxial cable, and thinner and longer coaxial cable will further increase the
series resistance of the mesh of outer conductors. This will reduce the effi
ciency somewhat, but it can be regained, if necessary, by further increasing the
number of turns. The longer coaxial lead of such an equaliser has a greater
parallel capacitance and series inductance, and this might be undesirable in
some leads. For example, it would increase the cable correction of a defining
lead (see section 5.4). In other meshes of a network, such increasedefficiency
equalisers are often a good choice.
(ii) It is possible to achieve nearperfect current equalisation by manually bal
ancing the flux in an equaliser core to zero, using the circuit and procedure
described in testing equaliser efficiency in section 3.2.2.
(iii) As an extension of this technique, the balancing can be done automatically, as
showninFigure 3.5, byapplyingfeedbackwithanoperational amplifier tocreate
a socalled active current equaliser [3]. This device employs two cores, one
mounted on the top of the other, and the coaxial lead is wound about ten times
around both cores. To simplify the figure, the coaxial leads and the single con
ductors are shown threaded only once through each core. A singleconductor
winding on one of the cores senses the induced voltage resulting from the mag
netic flux caused by the net current in the coaxial lead. An operational amplifier
amplifies and converts this voltage to a current that is passed through a single
conductor winding on the second core in such a sense as to null the net current
together with the voltage it causes in the first core (see Figure 3.6). This feedback
circuit increases the equaliser efficiencybythe voltagetocurrent gain. This gain
is the ratio of the feedback resistor R
2
to the input resistor R
1
multiplied by the
ratio of turns of the voltagedetection and the currentinjection windings:
Gain
R
2
R
1
N
det
N
inj
3:6
The concept of a lowfrequency coaxial network 55
Typical values are R
2
1 MW, R
1
1 kW, N
det
10 and N
inj
50, corres
ponding to a gain of 200. To derive this equation, it has been assumed that the
impedance of the detection winding is small compared to the input resistor, and
then the gain is practically frequency independent. Above a certain limit frequency,
this assumption is no longer valid and the gain decreases with frequency.
Furthermore, the gain should not be too large (in practice not larger than approxi
mately 1000), otherwise the active equalisers in a bridge network begin to oscillate
and generate excess noise. The more active equalisers that are used in a bridge
network, the smaller is the maximum gain allowing stable operation. A suitable
gain is of the order of 200500.
Care must be taken that stray capacitive currents are not coupled from the
outside into the network of inner conductors via the power supply leads to the
operational amplifier. To avoid this, active equalisers are often powered by bat
teries contained within the shielding box housing the device, but it must then be
ascertained that the device has been switched on and that the batteries remain in
good order for the duration of critical measurements. A better solution is to use a
totally isolating power supply as described in section 1.1.5.
Often, passive equalisers of enhanced efficiency are adequate if the operation
of the equalisers is assessed and a correction made for their imperfect efficiency, as
described in section 3.2.2. A core wound with a bifilar winding of ordinary con
ductors where one winding is connected into the inner circuit and the other into the
+
1
Y
and the righthand mesh, U
0
I/Y, hence
U U
0
1
YZ
2
3:8
In contrast to the original paper of Cutkosky [4], our representation of a cable
has the impedance Z divided in equal halves at the internal point A, as drawn in
Figure 3.12c. This gives rise to the factor of 2 in our expression where Cutkosky
has unity. We adopt our representation to simplify the cases when the clouds are
connected with two cables in series. The cables have admittances Y
1
, Y
2
and
impedances Z
1
, Z
2
, respectively, either because the lengths of similar cables are
different or because the cables have different characteristic impedances (see
Figure 3.15).
Y
1
Y
2
I'
I
I
1
I
Z
1
/2 Z
1
/2 Z
2
/2 Z
2
/2
U = 0
I' I
1
I
1
I
Figure 3.15 The combined effect of two cables in series
Traversing the lefthand mesh,
I
0
Z
1
2
I
0
I
1
Y
1
0
and traversing the central mesh,
I
0
I
1
Y
1
I
1
Z
1
Z
2
2
I
1
I
Y
2
0
Since YZ 10
6
for moderate lengths of the order of a metre of cable at audio
frequencies, we may neglect the secondorder terms, giving, on eliminating I
1
,
I I
0
1
Z
1
Y
1
2Z
1
Y
2
Z
2
Y
2
2
3:9
an expression which is equivalent to (3.7) for a single cable.
66 Coaxial electrical circuits for interferencefree measurements
The correction for potentialtransferring cables in series is identical.
If the two cables have the same characteristic impedance, then Z
1
/Y
1
Z
2
/Y
2
,
and (3.9) can be written as
I I
0
1
Z
1
Z
2
Y
1
Y
2
2
3:10
On comparing this result with (3.9) above, as expected, we see that the com
bined cable behaves as if it had total lumped impedance Z
1
Z
2
and admittance
Y
1
Y
2
. These total lumped impedances are what, to a sufficient approximation,
would be obtained by a direct measurement. Therefore, if two cables having dif
fering inductances and capacitances per metre have to be used, as for instance in
making connections inside and outside a cryostat, it is nevertheless a considerable
simplification if they have the same characteristic impedance and can be treated as
one cable for purposes of calculating a cable correction.
It also follows from the above that the effect of two cables in series, even if
they have the same admittance and impedance per unit length, cannot be directly
added. Combination is nonlinear because the effect of each cable is proportional to
the square of its length.
3.2.7 Triaxial cable
The results of the previous section can be modified by replacing a coaxial cable
with a triaxial cable where the intermediate shield conductor can be driven at a
remote end B by a voltage source U
guard
adjusted to raise the potential of the
intermediate shield to that of the inner conductor (see Figure 3.16).
D
A
U
B
U
guard
Z
int
I = 0
I = 0
U = 0
Figure 3.16 A triaxial cable with the intermediate shield raised to the same
potential as the inner conductor
The immediate obvious advantage is that the loading of the cable on a voltage
source U of high internal impedance Z
int
is removed. Furthermore, if the cable is
used for remote voltage sensing and U
guard
finely adjusted to null a current detec
tion transformer (see section 7.3.5) at the remote end B, the cable correction for
shunt admittance between inner and intermediate shield acting with the series
impedance of the cable is the same as calculated in the previous section, but the
sign is reversed. A similar conclusion follows if the cable is used to measure the
The concept of a lowfrequency coaxial network 67
current I in the inner at the remote end when a voltage detector between inner and
shield is nulled there. In either case, the effect is to create a defining cable that has
moved the plane where a voltage difference or a current is defined from A to a
remote point B.
The technique of employing triaxial cable to make fourterminalpair AC
measurements of quantum Hall impedance has been described by Ricketts [6]. This
technique is an alternative to multipleseries connections (see section 6.7.1) but
involves extra auxiliary balances and is hence not so elegant. Cabiati and DEmilio
[5] have extended this treatment to multipleshield cables.
References
1. A Guide to Measuring Direct and Alternating Current and Voltage Below
1 MHz. London: The Institute of Measurement and Control; 2003. pp. 7984
2. Kibble B.P., Robinson I.A. Guidance on Eliminating Interference from
Sensitive Electrical Circuits. NPL report DES 129. Available from http://
publications.npl.co.uk/ (search under publication)
3. Homan D.N. Applications of coaxial chokes to AC bridge circuits. J. Res.
NBS C. 1968;72C:16165
4. Cutkosky R.D. Four terminalpair networks as precision admittance and
impedance standards. Trans. IEEE Commun. Electron. 1964;83:1922
5. Cabiati F., DEmilio S. Low frequency transmission errors in multicoaxial
cables and fourport admittance standard definition. Alta Freq. 1975;44:60916
6. Ricketts B.W., Fiander J.R., Johnson H.L., Small G.W. IEEE Trans. Instrum.
Meas. 2003;52(2):57983
68 Coaxial electrical circuits for interferencefree measurements
Chapter 4
Impedance measurement
4.1 Improvements in defining what is to be observed
or measured
The standards compared by coaxial networks must be provided with terminal pairs.
The physical form of a terminal pair is a coaxial connector. Its inner contact is the
terminal connected to the component, and the outer contact is the terminal con
nected to its enclosing conducting case. A coaxial connector, which is a terminal
pair, is often called a port.
Unfortunately, the majority of lowfrequency reactance standards in laboratories,
and which are still being manufactured, are not provided with coaxial connections. It is
usually a simple matter, however, to replace the existing terminals with coaxial con
nectors and, where necessary, to provide a new outer conducting container.
It is very much the purpose of this book to encourage coaxial techniques within
the formalism of coaxial networks, to show that these techniques have the great
advantage of obtaining accurate and certain results, and to show how to implement
each concept with practical, constructable apparatus.
The bridge networks with which we are concerned will therefore consist of
impedances provided with, and defined in terms of, terminalpair coaxial terminals and
will be connected with conductorpair coaxial cables. Equalising the current in the
inner conductors and screening outer conductors of cables so that the currents are equal
and opposite will ensure that the cables have negligible external fields (see section
1.1.1), and therefore, the cables do not interact with each other so that their routing is
immaterial. The network will also have little response to external interference.
In general, the techniques we describe are appropriate for frequencies from
10 Hz to at least 1 MHz. Highest accuracy is attained for frequencies from about
400 Hz to 2 kHz.
At first sight, passive conductorpair networks seem unrelated to the design of
electronic circuits and equipment, but because of its emphasis on a complete
understanding of circuits and their interactions, AC bridges are a valuable subject
to study. The equalised conductorpair concept is also a powerful design tool for
electronic circuits, although somewhat different means such as the use of differ
ential input amplifiers are employed to achieve equalisation or minimisation of
loop currents. The role of power supply conductors and their associated current
return conductor (earth or ground) to and from a packaged circuit needs careful
consideration in relation to the circuitry as a whole.
Magnetically operated switches (relays) are often required in the inner circuit of
a coaxial system, for example, in designing a system for automated operation. The
switch contacts of these should be properly isolated by screens from the energising
coil, but inexpensive commercial components having this property are rare.
4.1.1 Ratio devices
Ratio devices which are voltage or current ratio transformers, are so constructed
that almost all of the magnetic flux threading one winding also threads another.
Their use produces more accurate and convenient bridge networks than the four
arm bridges of the older literature. Good flux linkage is accomplished by an
appropriate geometry of the windings, which are usually wound around a high
permeability toroidal magnetic core. Then the ratio of induced voltages or currents
in the two windings can be to within a part in a million or so, simply the ratio of the
number of turns of the windings. The small departures from the nominal turns ratio
can be measured and allowed for in high accuracy work. Ratio devices should be
subject to the same kind of precise electrical defining conditions as impedances
(see section 5.3). The zerovoltage port should have no current flowing through a
short, that connects the tap to the circuit of the outer conductors; that is, I 0 and
U 0 at this port. Other ports that have defined voltages across them should have
zero current flow through these ports. Ways of fulfilling these defining conditions
will be discussed in section 8.1.
4.1.2 Impedance standards
We have seen in section 1.1.1 the need to construct standards that have no sig
nificant electric or magnetic fields outside their containers. They must be provided
with terminalpairs which are coaxial connectors; the inner connector of a coaxial
connector is one terminal, and the outer connector the other terminal, of a pair. The
value of the standard is strictly defined by voltage differences across one or more of
the terminal pairs, and a current flow out of the inner terminal of another terminal
pair, with an equal as possible current flowing back through the outer terminal.
These specified defining conditions may be departed from, provided the departure
can be shown to have no effect on the measured value. For work of the highest
accuracy, the conditions at a single terminal pair may not be sufficiently repro
ducible, chiefly because of variations in contact resistance at the mating con
nectors. It is then necessary to establish internal defining points in the interior of
the device and to provide two additional terminalpairs which are coaxial connec
tions. The resulting four terminal pairs are analogous to the four terminals of the
familiar fourterminal DC resistor. Other definitions of standards are encountered,
but they can be viewed as degenerate examples of the two cases above.
Just as with lowvalued resistors measured with DC, where it is usually unsa
tisfactory for a standard to be provided with only two terminals because of uncer
tain contact resistance, and separate current and potential terminals are provided to
make a fourterminal resistance standard, a standard having four terminal pairs or
ports is required for AC work of the highest accuracy. The electrical conditions at
70 Coaxial electrical circuits for interferencefree measurements
each terminal pair (such as the voltage between inner and outer contacts when
zero current flows through either) must be precisely defined. But just as a two
terminal construction may be adequate for DC highvalued resistors, a simpler
twoterminalpair standard may serve for some AC measurements.
The concept of a direct impedance or admittance is very important. In general,
a component of admittance 1/Z will have admittances Y
1
and Y
2
from both its high
side and its low side, respectively, to the surroundings or to a current return
conductor, which, in the present context, is the outer shielding conductor of its
container and of coaxial cables connected to it. These admittances and the currents
flowing through them are distinct from the admittance of the component itself and
the current flowing directly through it. If I
1
and I
2
flow to the same conductor, all
three admittances form a D network, as shown in Figure 4.1 (see Appendix 1). We
are often concerned only with the ratio of the current I flowing directly through the
component to the voltage U across it, and this ratio is termed the direct admittance
of the component. Its reciprocal is the direct impedance of the component. A
network to measure the direct impedance Z must have a means of ignoring the shunt
currents I
1
and I
2
. For example, if the left side of Z is supplied from a voltage source
of negligible output impedance, the extra current I
1
drawn from it will not sig
nificantly affect its output voltage. If the right side of Z is connected to a nulled
detector, there will be no significant voltage across Y
2
, and consequently, I
2
will be
zero. Often, to obtain an equivalent situation in practice, auxiliary networks or bal
ances are employed, which remove the effects of I
1
and I
2
.
Z
U
I
Y
1
I
1
I
2
Y
2
a b
Figure 4.1 The direct admittance or impedance of a component
Section 5.3 is devoted to a careful treatment of the electrical definition of the
values of standards and ratio transformers, and we must emphasise that this is most
certainly essential. Once the defining conditions of standards are settled, the design
of proper comparison networks will follow naturally. Neglecting the definition
stage will, as in any other accurate measurement, inevitably lead to failure.
Standards can be represented by an equivalent circuit of lumped inductances,
capacitances and resistances connected into a network. An equivalent network is
such that its behaviour cannot be distinguished from the behaviour of the actual
device by making any kind of measurement at the terminals of the device over the
relevant frequency range.
Another minor advance in technique is that in most cases, the variable or
decade switched impedances used in older bridge technology can be replaced with
advantage by components of fixed value connected to a variable voltage generated
by a multidecade ratio transformer called an inductive voltage divider (IVD).
Impedance measurement 71
It is often useful to shunt an impedance with one of a much higher value so that
the impedance of the combination has exactly the required value. This is straight
forward if extra connections are made to the internal defining points, as shown in
Figure 4.2, but unfortunately, standards are not usually constructed in this way.
Z C
H
C
L
P
H
P
L
Z
Sh
Figure 4.2 Shunting a fourterminal resistor via extra connections to its internal
defining points
To shunt accurately in the absence of these extra connections is not quite
straightforward, as we will illustrate by considering shunting the fourterminal DC
resistor shown in Figure 4.3.
R
C
H
C
L
P
L
P
H
R
Sh
r
2
r
1
i
I
Figure 4.3 Shunting a fourterminal resistor (R
Sh
) R)
Because a small current flows through the shunting resistor, the defining
condition of zero current through the potential terminals P of the resistor being
shunted is not met. The voltage between the potential terminals is decreased from
the value
U I
1
R
1
R
Sh
r
1
r
2
1
4:1
(where I is the total current through the combination) which would be measured
between the internal defining points to
U ir
1
r
2
U
Ur
1
r
2
R
Sh
r
1
r
2
% U 1
r
1
r
2
R
Sh
4:2
between the potential terminals, where r
1
and r
2
are the resistances of the high and
lowpotential leads, respectively, and i is the small current I flowing through the
shunt. That is, the measured value of the shunted resistance is less than the value
that would be found between the internal defining points, which is correctly cal
culated from (4.1), by approximately the fraction (r
1
r
2
)/R
Sh
.
Asimilar decrease occurs if R
Sh
is connected across the current terminals instead.
72 Coaxial electrical circuits for interferencefree measurements
The same effect occurs if a fourterminalpair impedance is shunted, and this is
discussed in section 8.1.6.
The problem is avoided if the impedance has extra connections to the internal
defining points, as shown in Figure 4.2.
Very high value impedances can be made from more convenient lowervalued
ones by connecting them as a Tnetwork as shown in Figure 4.4. Their direct impe
dance Z
D
(section 4.1.2) can be calculated from a TD transformation (Appendix 1).
Z
D
Z
1
Z
3
Z
1
Z
3
Z
2
4:3
Z
1
I
in
I
out
U
out
U
in
Z
3
Z
2
Figure 4.4 A direct very high impedance constructed from a Tnetwork of lower
value impedances
Typical applications are the production of lowvalued capacitances when Z
1
,
Z
2
and Z
3
are all moderatevalued capacitors, and highvalued resistors when they
are all moderatevalued resistors (see section 6.4.10).
The operation of the network can be understood by viewing Z
1
and Z
2
as
forming a potential divider to supply a lower voltage to Z
3
so that the current
leaving Z
3
is attenuated. The direct impedance is the ratio of the voltage across
Z
1
Z
2
to this current and is consequently increased.
4.1.3 Formal representation of circuit diagrams and components
We, here, introduce the formal representation we use for circuit elements. The
formal representations of some particular devices are introduced when these devices
are described as they occur throughout the book.
A coaxial cable is represented as shown in Figure 4.5b. Thicker lines denote the
outer conductors of cables and suggest their lower resistance and impedance. The
cables connect at junctions with other cables and components by mating coaxial plugs
and sockets. A junction, which is pictorially represented in Figure 4.5a, will be for
mally drawn as in Figure 4.5b. The circles emphasise that the mating contacts of both
inner and outer conductors constitute terminals in the same sense as the single screw
terminals of older bridge practice. In choosing a particular kind of coaxial connector,
it is important that their mating surfaces make a reliable lowresistance contact.
(a)
(b)
Figure 4.5 (a) A typical coaxial connector. (b) Its formal representation
Impedance measurement 73
In general, the symbol
*
represents a singleconductor terminal or junction
point to which additional external connections can be made, and the symbol
represents a junction point, often internal to a device, to which additional connec
tions cannot be made.
Since the frequencies we are concerned with are less than radiofrequencies,
impedance matching of cables and connectors is unimportant.
An immittance, represented diagrammatically as the first diagram in Figure 4.9d,
may be either
(i) an impedance (Z) having reactive (jwX) and resistive (R) components or
(ii) an admittance (Y) having susceptive (jwB) and conductive (G) components.
Where the details of a general network are not of concern, the network will be repre
sented as shown in Figure 4.6 as a cloud having one or more coaxial connections.
Figure 4.6 A network represented as a cloud
The familiar symbols of electrical circuit theory represent resistors, capacitors
and inductors, which comprise the lumped equivalent circuit representation of
actual physical components; they will, in general, be surrounded by a screening and
conducting case to which the outer connectors of coaxial cables are joined, and so a
typical twoterminalpair component can be represented as shown in Figure 5.12a.
There is perhaps an implication that the component and its enclosing shielding
box also have coaxial geometry; this would be so in an ideal world, but the lim
itations of actual practical manufacture do not often permit it. The external fields
generated by a noncoaxial geometry of a component can be eliminated if neces
sary by enclosing it in a conducting magnetic shield.
Conducting surfaces forming electrical screens are represented by broken lines.
Where single conductors are necessary, they will be represented as curved
flowing lines with a loop to suggest their unpaired and unbalanced nature in that
currents in them give rise to surrounding magnetic fields.
Transformers need special attention, both because they are principal components
in most of the networks described and because their constructional topology is crucial
to their accuracy. All of the transformers we shall be considering have a toroidal core
made from highpermeability material. A toroidal form is preferred because of its
circular symmetry so that if it is excited by a uniform winding, very little magnetic
field exists external to the wound toroid. Some provision should be made for cancel
ling the effect of the advancing winding, which is equivalent to a single annular turn
around the toroid and can be approximately cancelled by a returning turn as shown in
Figure 7.17. See section 7.2 for actual constructional techniques.
74 Coaxial electrical circuits for interferencefree measurements
The windings of the transformer have to thread through the central space as
each turn is made, as shown in Figure 4.7a. To show this diagrammatically, as
shown in Figure 4.7b, the following convention will be used. First, on the left is the
representation of the core as a heavy solid line; next to this is one winding shown as
a spiral, and if this is completed by making a reverse annular turn next to the toroid
as shown in Figure 4.7a, its return is shown as a straight line parallel to the core.
The ends of the winding terminate on the inner and outer of a coaxial socket. In the
more elaborate example drawn in Figure 4.8, there is an electrical screen over the
first winding, shown as a broken line. This is followed by a second winding, and a
final overall electrical screening enclosure is shown. It is to be understood that a
given winding or screen threads all the windings or cores drawn to the left of it.
Other specialised or more elaborate transformers will be depicted similarly
the order in which the component parts are shown in the diagram follows the order
in which the transformer is assembled, starting with the core and working outwards
through the windings and shields.
Windings applied simultaneously side by side or, for example, as twisted
conductors are drawn above one another as shown in Figure 7.24.
Sometimes it is convenient to draw an assembly from right to left or from top
to bottom or from bottom to top instead of from left to right, as given earlier, and to
leave the context to clarify the intended construction.
We will always represent real bridge sources as voltage sources (i.e. sources of
negligible series internal impedance) and add the lumped representation of the
internal impedance of an actual source, if of any importance to the network in
which it occurs. This is not usually the case, and so the internal impedance will then
be omitted. Thus, an oscillator and its output coaxial terminations will be drawn as
shown in Figure 4.9h. We will initially ignore the extra complication brought about
(a) (b)
Figure 4.7 (a) A wound toroid. (b) Diagrammatic representation of a wound toroid
Figure 4.8 A toroid with two windings and a toroidal screen between them
Impedance measurement 75
(a) coaxial lead
(b) external terminal
(d) immittances (impedance,
capacitance, inductance)
(e) toroidal core
with a winding
(f) current equaliser
(g) coaxial immittance
component common diagrammatic representations
(c) internal junction point
(h) coaxial voltage source
(j) isolated voltage detector
(i) coaxial current source
(k) lockin amplifier
(with or without
preamplifier)
(l) batteryoperated
preamplifier
(m) preamplifier with
external supply
(n) mains earth safety wire
output
reference output
input D
output
output
input
output
input output
supply input
+
_
input output
input
reference input
D
input D
Figure 4.9 Diagrammatic representations of components
76 Coaxial electrical circuits for interferencefree measurements
by connections to the mains supply, which are either direct or indirect via the inter
winding capacitance of supply transformers in sources or detectors. Isolation can
actually be accomplished reasonably well by proper design of a screened transformer
(see section 1.1.5). We discuss the attainment of isolation in sections 1.1.41.1.10,
but for the present, it will be convenient to assume its accomplishment. For example,
we assume that all active components (generators, detectors, amplifiers, etc.) are
powered by batteries wholly contained within the screening case of the instrument.
In Figure 4.9, we tabulate the principal diagrammatic representations of the
components occurring in other diagrams in this book. For want of a better symbol,
we have represented a mains safety conductor as shown in Figure 4.9h, but this is
the only meaning of this symbol. It is not to be interpreted as, for example, a
connection to a chassis, an instrument case, a water tap or a ground plane.
The source and detector for a bridge network will consist of an isolated box
provided with a coaxial input connector. An isolated detector is represented as
in Figure 4.9j. Information on using a nonisolated detector will be found in
section 1.1.2.
We end with two particularly common components occurring in the construc
tion of impedancemeasuring networks (bridges). Figure 4.10a shows how the
important isolation (see section 1.1.5) of a source represented diagrammatically as
shown in Figure 4.10b can be accomplished in practice. Figure 4.11 illustrates the
functions of inline injection/detection transformers (see section 7.3.3), which can
either inject a small voltage or detect the presence of a current in an inner conductor.
U
(b) (a)
Figure 4.10 (a) An actual construction of an isolated source. (b) Its schematic
representation
U
I
(a)
(b)
Figure 4.11 (a) A voltage DU injected into an inner conductor. (b) Detection of a
current I in an inner conductor
Impedance measurement 77
Chapter 5
General principles of accurate impedance
measurement
We now begin our discussion of balanced current coaxial techniques as applied to
impedance measurement with AC bridges. As mentioned in the introduction, this
provides a good example of how these techniques can be applied in general to
sensitive electrical measurements of other parameters voltage, current, power and
so forth.
Before moving on, it is as well to note the advantage of a substitution mea
surement in comparing two nearly equal impedances, and indeed in metrology in
general. We can make a formal statement of the philosophy of a substitution
measurement as follows.
If two devices to be compared are so similar that the measuring system does
not respond to minor differences in their construction, then the small difference
in their values can be measured by substitution without incurring any systematic
(type B) error arising from the measuring system.
For example, if two weights X and Y of similar value are compared by placing
each in turn on the righthand pan of a balance, which has a third, similar weight on
the lefthand pan, by adding small makeweights to the righthand pan to achieve
the two separate balances for X and Y, the difference in the values of the makeweights
will reflect the difference between X and Y irrespective of whether the balance arms
are of equal length or whether the balance indicates zero with no weight on either pan.
In the present context, no significant type B (systematic) error will be incurred
for a substitution measurement of two similar direct impedances Z
1
and Z
2
if the
measuring system (bridge) is insensitive to any difference between the shunt
impedances at their terminal ports.
5.1 The evolution of a coaxial bridge
We begin with a discussion of the fourarm Wheatstone bridge and show how, after
it has been converted to a coaxial circuit, the measurement can be made to respond
only to the direct admittances or impedances (see section 5.3.2) of the four com
ponents in the bridge arms.
The reader will probably be familiar with the Wheatstones bridge network
generalised for AC comparison of twoterminal impedances, as drawn in Figure 5.1.
We assume an isolated source (see section 1.1.5) and an isolated detector (e.g. one
powered by batteries contained within its complete screening case) in order to
facilitate the following discussion.
The balance of this bridge made by adjusting Z
3
will be more or less affected
by admittances (usually stray capacitances) between each corner of the bridge and
its surroundings, and measurements made with it will be erroneous.
These admittances can at least be made much more definite by enclosing the
components and the conductors in a complete screen, as shown in Figure 5.2.
If the network were to be physically arranged somewhat as in the diagram, the
mutual inductive coupling (see section 1.1.1) between its three meshes could cause
very significant error, particularly if the impedances involved are small. The circuit
layout can be arranged so that conductors carrying equal and opposite currents are
adjacent, to minimise inductive coupling, but at the expense of increased capacitive
coupling.
Alternatively, current equalisers added to each mesh as shown will ensure
equalised currents in the screen, so that the network is properly coaxial, and the
physical layout should be immaterial.
D
U
a
U
b
Z
1
Z
2
Z
3
Z
4
Figure 5.1 A singleconductor AC Wheatstones bridge
D
Y
S1
Y
S2
Y
D2
Y
D1
Z
3 Z
1
Z
2
Z
4
Figure 5.2 A completely screened singleconductor AC Wheatstones bridge
80 Coaxial electrical circuits for interferencefree measurements
The network will now be immune to external interfering electric and magnetic
fields, but probably the admittances (in practice, lossy capacitances) Y
S1
, Y
S2
, Y
D1
and Y
D2
from the corners of the bridge will be increased. Y
S1
, Y
S2
, Y
D1
and Y
D2
and
the cable capacitances are the adjustable circuit components rather than indefinite
stray capacitances.
To facilitate discussion, the network has been redrawn as shown in Figure 5.3,
where the extreme righthand and lower lines represent the outer conductors of the
cables and components. Y
S1
, Y
S2
, Y
D1
and Y
D2
will, in general, cause large errors,
but their values can be adjusted by the following procedure so that they cause no
significant error.
D
a
b
c d
Y
S1
Y
S2
Z
3 Z
1
Z
2
Z
4
Y
D2
Y
D1
Figure 5.3 A simplified line drawing of a completely screened Wheatstone bridge
The main balance having been made by adjusting Z
3
to null the detector D, the
bridge nodes c and d can be brought to the same potential as the outer conductors
by adjusting the ratio Y
S1
/Y
S2
to null D with Y
D1
temporarily shorted. It will
probably be necessary to iterate the main balance with this auxiliary balance until
both are simultaneously obtained. When this is true, Y
D1
and Y
D2
have no voltage
across them and so do not affect the main balance whatever their value. The
adjustment of the Y
S1
/Y
S2
ratio is called a Wagner balance in bridge literature, and
there are many examples later in this book.
If now the Wagner balance is temporarily misadjusted by an amount that can
later be reset, c and d will no longer be at the same potential as the screen, but they
can, nevertheless, be set to be at the same potential as each other by adjusting the
ratio Y
D1
/Y
D2
until D is again nulled. This adjustment can be termed a detector
network auxiliary balance. The bridge balance is now immune to the values of Y
S1
and Y
S2
. If their temporary misadjustment is now restored, the bridge balance is
immune to both auxiliary balances to an extent that is the product of the precision
with which either auxiliary balance has been made. For example, if each has been
made with a precision such that they individually do not affect the main balance by
more than 1 in 10
3
, their combination will not affect the main balance by more than
1 in 10
6
. Also, Y
D1
, Y
D2
, Y
S1
and Y
S2
need only 1 in 10
3
stability in contrast to the 1
in 10
6
stability of the principle impedances Z
1
, Z
2
, Z
3
and Z
4
. An example of the
application of this philosophy in slightly different circumstances can be found in
the quadrature bridge described in sections 9.2.19.2.3.
General principles of accurate impedance measurement 81
This simple fourarm bridge still has practical application for impedance
measurement at higher frequencies at which it may be difficult to make satisfactory
ratio transformers.
More often in the bridge designs described later in this book, the components
Z
1
and Z
2
are replaced by the outputs of the windings of a voltage ratio transformer.
Because the ratio of the output voltages is very little affected by loading the
transformer outputs with shunt admittance, a detector network auxiliary balance is
unnecessary. The node c is connected to the screen by a shortcircuit, which can be
temporarily removed, and Wagner components Y
S1
and Y
S2
can be adjusted so that
the main bridge balance is unchanged whether the short is in place or not. When
this condition is achieved, there will be no current through the short and no voltage
across it, and this constitutes one of the defining conditions of the voltage ratio
transformer. In the following section, we illustrate this principle by applying it in a
simple bridge.
5.1.1 A simple coaxial bridge as an example of a coaxial network
Figure 5.4 is a partly pictorial section of a simple coaxial capacitance comparison
bridge. It will enable the reader to appreciate the construction of a toroidal trans
former having two screens between the windings. Figure 5.5 shows the schematic
representation of the same bridge. The transformer construction (see section 1.1.5)
D
Figure 5.4 A pictorial representation of a simple capacitance ratio bridge based
on a voltage ratio transformer
D
C
1
C
2
+NU
U
Figure 5.5 A diagrammatic representation of the bridge represented in Figure 5.4
82 Coaxial electrical circuits for interferencefree measurements
ensures isolation between its primary and secondary windings, and transformers of
this kind are used to isolate a commercial source from a measuring network.
Very accurate impedance comparison with a bridge does not need separate
measurement of voltage and current; it is sufficient that the same current flows
through both the devices to be compared. This is so in this simple coaxial bridge if
the detector is nulled, that is, it draws zero current from the mesh. The ratio of the
impedances is then the ratio of the voltages across them, and precise voltage ratios
can be obtained from a properly constructed transformer. In the older bridge liter
ature, the voltage ratio is provided by the potential drops across the known ratio of
two other impedances in series, and Figure 5.6 shows how this bridge relates to the
fourarm Wheatstones bridge discussed in the preceding section.
I = 0
I = 0
D
D
+NU
U
Z
3
C
1
C
2
Z
1
Z
2
Z
4
Y
W1
Y
W1
Y
W2
Y
W2
(a)
(b)
Figure 5.6 The relationship between (a) a fourarm bridge and (b) one based on a
N:1 voltage ratio transformer
A properly designed voltage ratio transformer has a very small mutual output
impedance of one winding with respect to the other. That is, if one output has its
voltage altered, for example, by a shunt load, the voltage across the other winding
General principles of accurate impedance measurement 83
will, to a very good approximation if it is constructed to have only a small leakage
inductance (see section 7.1.3), also alters so that the ratio of its output voltages is
unaltered. This ensures that the adjustment of the Wagner components Y
W1
and
Y
W2
, which have been added in Figure 5.6b, is less critical.
As we noted earlier, it is one of the defining conditions of this kind of voltage
ratio transformer that no significant current should flow through the short between
the tap of the windings and the outer circuit. In the previous section, we saw how
this can be achieved by temporarily removing this short and adjusting the Wagner
components to restore the bridge balance. Alternatively, the detector can tem
porarily be connected in place of the short and nulled.
5.2 The validity of lumped component representations
The flow of electrical energy driven by an alternating voltage through an actual
electrical component can be represented by considering the component to be a
network of some or all of the idealised components of inductance L (in which the
energy is associated only with a magnetic field), of capacitance C (in which elec
trical energy only exists in the electric field) and of resistance R (in which the
energy is dissipated entirely as heat and there are no associated electric or magnetic
fields).
In an actual component, all three processes take place together within the
device and in the surrounding space. Consider a coil of insulated wire carrying a
current. Each short section of the wire will be surrounded by a magnetic field so
that it possesses inductance, will dissipate energy through its resistance and, as it is
at a different potential from its neighbouring sections, there will be electric fields
within and in the vicinity of the coil.
At sufficiently low frequencies, the magnetic fields from successive sections
will have the same phase, and the total field can therefore be treated as a single
entity as if it arose from an ideal inductance L. The effect of electric fields can be
modelled in a similar way as an ideal capacitance C. The total resistive dissipation
of the device can also be treated collectively as an ideal resistance in series with the
inductance in the network of ideal components drawn in Figure 5.7. This equivalent
circuit then behaves in exactly the same way as the actual component up to fre
quencies higher than the selfresonant frequency w
r
(LC)
1/2
because the self
resonant frequencies of the individual sections of the coil are even higher, and the
electric and magnetic fields from each add to give the total field of the device in a
way that still produces a constant mean phase throughout the volume of the coil.
C
R L
Figure 5.7 An equivalent circuit for a resistor
84 Coaxial electrical circuits for interferencefree measurements
Thus, representing the actual behaviour of the coil by lumped component para
meters is valid provided the wavelength of the oscillatory magnetic and electric
fields associated with the coil is long compared with its dimensions. The values of
the parameters may be frequency dependent.
If two devices are separated so that their electric and magnetic fields do not
share a common region of space, their combined behaviour, when connected
together, is represented by the connection of their respective ideal networks. This is
not the case if their fields partially overlap. For example, two equal inductors of
value L, when connected in series, have a combined inductance 2L when well
separated, but if they are brought together, their combined inductance approaches
one of the theoretical limits 4L or 0, depending on the sense of connection. In
general, therefore, it is the spatial distribution and interaction of the electric and
magnetic fields of a device with its conductors that should be considered when
deciding whether a particular lumped parameter representation is appropriate.
5.3 General principles applying to all impedance standards
5.3.1 The physical definition of a standard
The value of capacitance C or inductance L is defined by the energy e CU
2
/2 or
LI
2
/2 stored in its electric field E or magnetic field H created by the voltage U or
current I, respectively. These energies can be expressed as integrals over a volume
V in terms of the fields as 1/2
ee
0
E
2
dV or 1/2
mm
0
H
2
dV. It, therefore, follows that
any neighbouring object that modifies these electric or magnetic fields will alter the
value of the capacitance or inductance. A good capacitance or inductance standard
will be so constructed that no appreciable modification can occur. By the recipro
city principle explained in section 1.1.1, this will be so if the component is con
structed in such a way that it itself has no appreciable external fields. The result of a
satisfactory construction could be termed the physical definition of a standard.
Ideally, capacitors should be completely enclosed within a conducting screen,
which is continuous with the outer conductors of coaxial terminals. The capacitor
should be connected to the inner conductors so that it constitutes a two or four
terminalpair component (see sections 5.3.6 and 5.3.8). Its magnetic flux generated
by these connections and the displacement current within the capacitor is usually
negligible.
Inductors should be toroidally wound by one of the schemes of section 7.2.1,
and, since electric fields emanating from inductors of millihenry values or greater
are likely to be significant, they should be totally enclosed in a conducting screen
with coaxial terminals in the same way as a capacitor.
Astandard of resistance of value R should only dissipate energy at a rate WI
2
R;
the geometry of its conductors should be such that when carrying an alternating
current of frequency f, the stored energy per cycle of its electric or magnetic fields
(LI
2
CU
2
)/2 is small by comparison with the energy I
2
R/f dissipated in the resistance.
Like standards of capacitance or inductance, it should also be completely enclosed in a
conducting screen and have coaxial terminals.
General principles of accurate impedance measurement 85
Actual construction of commercially produced standards, so far, seldom even
approach these ideals, and so the manner in which interactions with external fields
are minimised during measurement needs to be clearly specified.
5.3.2 The electrical definition of a standard impedance
The considerations of this section apply equally to both AC and DC measurements.
For DC measurements, replace impedance and admittance by resistance and
conductance, respectively.
The electrical conductors of a measurement network are connected to the
terminals of standards. We will call the electrical definition of a standard impe
dance, the impedance defined as the ratio of the potential that exists between two
designated terminals to the current that flows through one designated terminal. The
admittance of a standard is the reciprocal of this ratio.
When an impedance is provided with separate potential and current
measuring terminals, the junctions at which conductors from the impedance divide
to go separately to the potential and current terminals are called internal defining
points. The junctions are not strictly points, being blobs of metal, but if current or
voltage measurements are made remotely from them at terminals along conductors
at a distance much greater than the crosssection of the conductors, the blobs
behave as linear networks and the current and potential paths act as if they meet at
an actual point within them. The reciprocity principle ensures that the location of
these points remains the same if the roles of current and potential conductors are
interchanged. The defined impedance is that which would be found by direct
measurement between these points if they could be accessed. The purpose of the
definitions below is to reproduce as closely as possible this defined impedance by
measurements actually made at the remote current and potential terminals.
Internal defining points are explicitly labelled a and b in Figure 5.10, which
represents a fourterminal component having separate current and potential
terminals.
Additional potential or current conditions may be imposed at the same or other
terminal points to increase the precision of a definition, and certain kinds of definition
are advantageous. In an actual measurement, it may well be that the defining condi
tions are not fulfilled exactly, but instead it should be demonstrated that the result of
the measurement is the same as if the defining conditions had been strictly fulfilled.
We emphasise again that the concept of a direct impedance or admittance (see
section 4.1.2) is important. With reference to Figure 5.8, the direct impedance of the
component is the ratio of the voltage U between its terminals a and b to the current I
flowing through one terminal. The currents I
1
and I
2
internal to the component
through various admittances Y
1
and Y
2
to its container or surroundings do not form
part of its direct impedance. They should be so routed, or reduced to zero, or their
effect removed by auxiliary balances, as to measure its direct impedance correctly.
The electrical definition of a standard, or defining conditions that can be
shown to be equivalent, must be strictly realised in a balanced bridge network if
accurate results are to be obtained. For example, consider the familiar case of a
86 Coaxial electrical circuits for interferencefree measurements
fourterminal DC resistor where the electrical definition is realised in practice by
the potentiometric method of comparison. In this method, the same current flows
through two resistors in series, and the voltage differences between their respec
tive potential terminals are compared under conditions of zero current flow
through these terminals. The same defined value is obtained from Kelvins double
bridge described in section 5.6, even though in this network, appreciable currents
flow through the adjacent potential terminals. The reason for the equivalence of
these conditions to those of the exact electrical definition is discussed in that
section.
5.3.3 Twoterminal definition
This definition applies when the component has only two terminals. Its impedance Z
is defined as the ratio of the voltage U between these terminals to the current I leaving
one of them, as shown in Figure 5.9. Since there are only two terminals, it is implicit in
this definition that capacitance and conductance to the surroundings are negligible and
that the current leaving the component is equal to the current flowing through it,
otherwise the standard is ill defined.
I
U
I
Z = U/I Z
Figure 5.9 A twoterminal definition of a component
This simple definition is electrically unsatisfactory in that the terminals are
only an approximation to internal defining points. Actual terminals are objects of
finite physical size having different potentials between different points at which
connections might be made to them. Moreover, uncertain potential differences arise
through uncertain contact resistance, mutual inductance and capacitance between
the measurement leads, and the terminals and conductors beyond them. Hence,
more elaborate defining conditions, which overcome these problems, are needed
for accurate standards.
Z
U
I
Y
1
I
1
I
2
Y
2
a b
Figure 5.8 The direct impedance or admittance of a standard
General principles of accurate impedance measurement 87
5.3.4 Fourterminal definition
This definition, illustrated in Figure 5.10, may already be familiar as that used for
accurate DC resistors because it eliminates any uncertainty caused by indetermi
nate terminal resistances. The standard is provided with two extra terminals con
nected to the internal defining points a and b between which the voltage difference
is defined. The usual way of arranging the four terminals is only suitable for DC or
for lowfrequency measurements of mediumvalue impedances because consi
derable electric and magnetic fields emanate from them and the conductors going
to them.
I
U
Z
a b
Z = U/I
I = 0 I = 0
Figure 5.10 A fourterminal definition of a component
5.3.5 Fourterminal coaxial definition
This unfamiliar definition, illustrated in Figure 5.11, is formally identical to that of
the previous section, but the physical arrangement of the terminals is altered to two
terminal pairs. The connecting cables can be properly current equalised, and hence,
cable corrections (see section 5.4.2) can be made to measurements. The compara
tively large capacitance between inner and outer conductors of each coaxial cable
shunts the resistor and gives it a phase angle. Consequently, this definition is only
suitable for resistors measured at lower frequencies. It is particularly useful for
resistance thermometers (see section 9.6).
I
U
I
Z
I = 0
I = 0
Z = U/I
Figure 5.11 A coaxial fourterminal definition of a component
5.3.6 Twoterminalpair definition
The more usual alternative arrangement of Figure 5.12 is more appropriate for
highervalued impedances. The admittances Y
mn
and the impedance z internal to the
impedance (Figure 5.13a) may be represented more simply as shown in Figure 5.13b
as just three lumped components. The impedance is again defined as the ratio of the
potential U
H
between the two terminals of one terminalpair to the current I
L
out of
and back into the terminals of the other terminalpair, with the additional condition
that there is zero potential between this terminalpair. The currents into and out
of the lefthand terminalpair are also equal since the total current leaving the
88 Coaxial electrical circuits for interferencefree measurements
righthand terminalpair is zero, but in general, these currents will not equal I
because of internal admittances Y
H
and Y
L
to the screen whose actual value plays no
part in the defined direct impedance of Z (see section 4.1.2). This is because U
H
is
the potential difference measured across the lefthand terminalpair, irrespective of
how much current flows through Y
H
, and since U 0 between the righthand
terminalpair, no part of the current I
L
is shunted away through Y
L
. This is in
contrast to the previous section where Y
H
and Y
L
are cable capacitances added to
the measured direct admittance of the device.
The terminals on both the right and lefthand sides carry current, and the
potentials measured between them will be subject to contact resistance variations.
The total shunt admittance measured at the input terminalpair with the output
terminalpair shorted is Y
H
1/Z, and the similarly defined output shunt admittance
is Y
L
1/Z. The direct impedance Z z, which is the defined admittance to be
measured, is U
H
/I
L
when U
L
0.
In section 5.1, we discussed how the effects of shunt admittances on measuring
networks can be eliminated and conditions equivalent to this definition can be
achieved.
5.3.7 Threeterminal definition
Wherever this phrase is encountered, the intended meaning needs to be examined
with especial care. It is usually meant to be applied to a guarded (i.e. electrically
screened) twoterminal device.
Z
= U
H
/I
L
Z
I
H
I
L
U
H
U
L
= 0
Figure 5.12 A twoterminalpair definition of a component
Z
z
1 2
3 4
Z
= U
H
/I
L
z
U
L
= 0
I
L
I
L
I
H
I
H
U
H
Y
23
Y
13
Y
24
Y
14
(a)
Z
= U
H
/I
L
z
I
H
I
L
I
L
I
H
Y
H
Y
L
U
H
Z
z
(b)
U
L
= 0
Figure 5.13 (a) The internal admittances of a twoterminalpair component.
(b) Simplified equivalent circuit
General principles of accurate impedance measurement 89
With reference to Figure 5.14a, the screen current I
G
, which flows through the
measurement network, should play no part in the definition of the component.
Impedances defined in this way cannot be accurately measured in a current
equalised network. Highvalued resistors are often provided with three single
terminals, and their value is intended to be realised in accordance with the above
definition. Some AC capacitors are also defined in this way, and despite being
provided with two coaxial fittings, they cannot be used in a currentequalised net
work as the outer conductor of the highpotential port serves only as a termination
of the screening of the highpotential lead and is not connected to the outer of the
lowpotential port. The comparison of impedances defined in this way is discussed
in the older literature [1]; from our point of view, it is better to reconstruct and
redefine them as a two terminal pair, as shown in Figure 5.12.
5.3.8 Fourterminalpair definition
To overcome the causes of uncertainty associated with a twoterminalpair defini
tion, it is necessary to combine the concepts of four and twoterminalpair defi
nitions. The result, shown in Figure 5.15, is a fourterminalpair definition, which is
precise enough for all present practical applications and all values of impedances. It
is shown in section 5.4.4 that the uncertainty in the product of the impedances and
admittances associated with the coaxial cables and connectors used to connect the
impedance into a network sets a limit to how precisely this definition can be rea
lised, but this precision is more than adequate for all practical purposes.
The numbering of the terminal pairs or ports in Figure 5.15 follows that of the
original paper by Cutkosky [2], and this has become the standard notation. Alter
natively, and as we shall use in the rest of this book, they may be denoted as current
(C) and potential (P) ports at the Hi and Lo ends of the component and labelled
C
H
, P
H
, C
L
and P
L
as shown.
A generator may be connected across C
H
to provide current through the
component, which can be compared with another two or fourterminalpair
impedance connected to it at C
L
.
Z
Z
U
U
(a)
(b)
I
H
I
H
I
L
I
L
I
G
Y
L
Y
H
I
YH
U
L
= 0
U
L
= 0
Z = U/I
L
Z = U/I
L
Figure 5.14 (a) The usual threeterminal definition of a component and (b) its
equivalent circuit
90 Coaxial electrical circuits for interferencefree measurements
The defining conditions for the device are not affected by the shunt impe
dances of cables or other components connected to C
H
or P
L
all that matters is
that the inward and outward currents at C
L
are identical and that the defining
conditions at P
H
and P
L
are fulfilled. We anticipate later discussion by noting that,
for example, P
H
could be connected to the known output potential across a port of a
transformer and a generator connected to C
H
could be adjusted until no current
flows at P
H
. The remaining defining conditions could be fulfilled by simultaneously
bringing a detector connected to P
L
to a null.
The fourterminalpair definition appears, at first sight, to be both somewhat
abstract and complicated, but it is easy to show that it has a close relationship to the
familiar fourterminal impedance (see section 5.3.4), which has separate current
and potential terminals.
Consider Figure 5.16 in which the impedance Z
0
and the inner conductors of the
coaxial cables are drawn as a fourterminal impedance. This impedance can be seen
as being in series with the small fourterminal impedance z formed by the outer
conductors. Because of the current equalisation, the same current I flows through
both, and because of the condition U 0 between inner and outer at P
L
, the potential
difference U at P
H
belongs to the sum of the inner and outer impedance. The actual
defined impedance is therefore U/I Z Z
0
z. The fourterminalpair definition is
Z
Z
z
1
2
3
4
(a)
(b)
I
H
I
H
I
H
I
L
I
L
I
L
I = 0
I = 0 I = 0
I = 0
C
H
C
H
P
H
P
H
P
L
P
L
C
L
C
L
U
H
U
H
Y
H
Y
L
U
L
= 0
U
L
= 0
Z = U
H
/I
L
Z' = Z + z = U
H
/I
L
Figure 5.15 (a) The fourterminalpair definition of a component. (b) Its
equivalent circuit
U
I
z
Z'
U = 0
Figure 5.16 A fourterminalpair component drawn as the sum of two four
terminal components in series
General principles of accurate impedance measurement 91
therefore a natural extension of the familiar fourterminal DC definition, which
produces a carefully defined impedance suitable for measurement in an equalised
current network.
With reference to Figure 5.15, the impedance Z U/I is fixed completely by
the defining conditions at the internal defining points. The conditions at P
H
and C
L
are slightly different from those at the internal defining points because they include
the effects of any internal cables between the internal defining points and these
ports. The effects of these cables add to those of external cables connecting
between P
H
and C
L
, and the measuring network and their combined parameters
give rise to Z
2
, Y
2
, Z
3
and Y
4
, as drawn in Figure 5.23. We will call the combination
of internal and external cables defining cables. The calculations of section 5.4.4
must take account of the internal cables, which are in series with the defining
cables. These conclusions are at variance with previous literature, which assumed
that the effect of any internal cables is negligible.
The exact point in the measuring network at which a defining cable ends is
either the defining plane of a detection transformer (section 7.3.5) or a bridge
Tjunction at which connection to a detector is made or the defining plane of
an injection transformer (section 7.3.4) which injects a voltage to balance a combi
ning network or where an impedance is switched in to make this balance (e.g. in
Figure 9.22).
Neither the internal impedance of the source at C
H
nor any load impedance
across P
L
matters if the defining conditions are satisfied.
5.3.9 Measuring fourterminalpair admittances in a twoterminal
pair bridge by extrapolation
If the defining conditions at a terminal pair (port) are not exactly fulfilled, we
can, nevertheless, deduce the balance condition of a bridge, which corresponds to
exact fulfilment by extrapolating the change produced on increasing the non
fulfilment by a known amount. This is an example of extrapolation techniques,
which make use of the property of linearity and which can produce an accurate
result as if the defining conditions were exactly fulfilled. We illustrate the principle
by applying it to measuring a fourterminalpair standard with a twoterminalpair
bridge, as shown in Figure 5.17.
We note the bridge balance readings in the original state and again when
known admittances have been added to C
H
and P
L
, C
0
H
and P
0
L
. Then an extra
polation can be made to the balance readings to calculate those that would have
been obtained under the correct defining conditions.
In the twoterminalpair bridge shown in Figure 5.17, P
H
is connected to an
output tap of a transformer via a defining cable and a detector to C
L
via another
defining cable. As usual, the current returning via the outer conductors is made
equal to that in the inner conductors by an equaliser, here shown in the defining
cable going to P
H
. The output taps of the transformer only produce their correct
output voltages when zero current is drawn from them (i.e. their defining conditions
are satisfied), and P
L
and P
0
L
should have zero potential across them. Extrapolations
92 Coaxial electrical circuits for interferencefree measurements
to these conditions are easily made by adding admittances to the open ports of each
fourterminalpair admittance Y and Y
0
. For Y, extrapolation to the defining con
dition of the output tap of the transformer can be made by connecting across C
H
an
extra admittance which, is for simplicity, equal to that measured between the inner
and outer conductors at C
H
with C
L
shorted and the defining cable disconnected
from the transformer output tap. Adding this extra admittance doubles the current
taken from the transformer output port while leaving the fourterminalpair admit
tance unaltered. Extrapolation back to zero admittance presented to the transformer
output port is therefore equivalent to achieving its correct I 0 defining condition.
The correct defining condition of I 0 at port P
H
of Y is still violated because of
current through the capacitance between its inner and outer conductors, which causes
voltage drop down the defining cable. This can be accounted for by applying a cable
correction as described in section 5.4.4, but note that the correction has the opposite
sign to the case where I 0 at P
H
.
A correction for the error caused by violation of the defining condition at P
L
can
similarly be found. When the bridge is balanced, the voltage U
D
between the inner and
outer conductors at the Tjunction where the detector is connected is zero, and no
current flows there. Instead, a small portion of the current through the defining point b
flows through the admittance from b to the detector Tjunction between the inner and
outer conductors and violates the defining conditions U 0 and zero current between
the inner and outer conductors at b. A second admittance Y
CL
, equal in value to that
measured at C
L
with P
H
shorted, when connected across P
L
will double the potential
there, and therefore, the current between inner and outer at b and extrapolation back to
zero admittance will reproduce the proper defining conditions at b. Again, a cable
correction with the opposite sign must be made for the defining cable at C
L
.
The same procedure applied to Y
0
will yield its correct defining conditions.
It is not necessary to use the particular values of shunt admittances suggested,
provided the values are known. Let the ratios of the shunt admittances added to C
H
and P
L
to those measured at P
H
and C
L
with C
L
or P
H
shorted be k
1
and k
2
,
respectively, and those added to C
0
H
and P
0
L
to those measured at P
0
H
and C
0
L
with C
0
L
D
Y
a
b
a'
b'
P
H
'
P
H
C
H
C
H
'
C
L
'
C
L
P
L
'
P
L
Y'
U
D
= 0
I = 0
I = 0
I 0
U 0
U U
0
U
0
Figure 5.17 A twoterminalpair bridge used to measure fourterminalpair
admittances
General principles of accurate impedance measurement 93
or P
0
H
shorted be k
0
1
and k
0
2
, respectively. The individual changes of bridge balances
brought about by adding such shunt admittances in turn or to the appropriate ports
can be weighted by k
1
, k
2
, k
0
1
and k
0
2
and added algebraically to the bridge balance to
give the total correction for the ill definition of Y.
Alternatively, if k
1
k
2
k
0
1
k
0
2
the shunt admittances can be added simul
taneously and the total correction obtained from just one extra bridge balance.
5.3.10 Adaptors to convert a two or fourterminal definition to a
fourterminalpair definition
There are many situations where it is required to convert the terminal configuration
of an existing device so that it can be measured as a fourterminalpair component,
with all the clarity of definition that it implies. It is instructive to examine the
topology of an adaptor to do this because this same topology is a good way to mount
an isolated component in a screening can and is employed in the construction of
highfrequency standards as described in section 6.5.712. The two problems to be
solved are, first, to bring the outer conductors of four cables together in such a way
that their junction possesses zero fourterminal impedance and, second, to continue
their inner conductors going to the component in such a way that there is acceptably
small mutual inductance and capacitance between these leads.
The outer conductors can be joined at a fourterminal zeroimpedance junction.
A fourterminal zero impedance has the property that the potential between its
potential terminals remains zero when a current flows between its current termi
nals. A zero DC resistance can have a conductor joining its current terminals and
another joining its potential terminals and a link of small crosssection between a
point on either conductor. Then, however much current flows between the current
terminals, no potential difference appears across the potential terminals. A zero AC
impedance must also have no significant mutual inductance between the current
and potential links, and this can be achieved by an orthogonal geometrical layout.
The inner conductors need to be continued as the inner conductors of sepa
rately screened and insulated cables twisted together in current and potential pairs.
A possible construction is illustrated in Figure 5.18a, where the outer conductors of
P
L
and P
H
are connected together at the crossshaped junction with the outer con
ductors of C
H
and C
L
. This orthogonal geometry minimises mutual inductance
between the current and potential paths. The adaptor is shown with spade termi
nations suitable for connecting to the bindingpost terminals of a fourterminal
component, to convert its definition to four terminal pair, but they could equally
well, for example, be connected to a single port, making a twoterminal connection
by connecting C
H
and P
H
to one terminal and C
L
and P
L
to the other, or to other
terminal arrangements [3]. If the component has a screening can, it should be
connected to the single conductor from the centre of the cross.
Figure 5.18b, c and d are rigid adaptors devised ad hoc to make the conversion
for particular designs of components. They might be sufficiently good for their
purpose but exhibit small mutual inductive interactions. Figure 5.18b and d also
have a small internal shunt capacitance.
Impedance standards for higher (microwave) frequencies are frequently single
port devices. They can be configured for fourterminalpair measurement with an
94 Coaxial electrical circuits for interferencefree measurements
adaptor consisting of a mating connector mounted on, but insulated from, the top of
a cylindrical or square conducting box of minimal size. The fourterminalpair
connectors are mounted 90 degrees apart at the same height on the sides of the box
to provide orthogonal current and potential connections as in Figure 6.26. Radial
connections go from the inner conductor of the mating connector to the low and
from the outer conductor to the high inner conductors of the fourterminalpair
connectors. The outer conductors of the fourterminalpair connectors are con
nected by the box. Shielding may need to be completed for interference elimination
by a shield connected to the box which encompasses the standard. The residual
fourterminalpair capacitance of the adaptor up to the mating plane of the mating
connector with no standard connected should be measured and subtracted.
5.4 The effect of cables connected to the ports of impedance
standards
We have seen that impedance and admittance standards, that is, resistors, inductors
and capacitors, possess inaccessible internal defining points where corresponding
current and voltagemeasuring conductors fromthe accessible terminals of the device
are connected together internally to the resistive, capacitive or inductive element. The
value of the standard is defined in terms of the current and voltage differences at these
internal defining points, and the effect of connecting cables between these points and
the measuring bridge or instrument must be accounted for in accurate measurements.
The following calculations build on the basic considerations of section 3.2.6.
We emphasise again that because cable corrections combine nonlinearly, the
correction for an individual cable added later cannot be directly applied. The
properties of all cables connected in series up to the internal defining points of an
(b)
(a)
(c) (d)
C
L
P
H
C
H
C
H
C
L
P
L
P
L
P
H
screen
Figure 5.18 (a) An adaptor to convert two and fourterminal components to a
fourterminalpair definition. (b, c and d) Rigid adaptors for
particular applications
General principles of accurate impedance measurement 95
impedance must be accounted for. This is because the effects of two cables in
series, even if they have equal admittance and impedance per unit length, cannot be
directly added because the effect of each cable is proportional to the square of its
length. Therefore, the properties of both cables must be known, and the effect of the
combination must be calculated (see section 3.2.6).
5.4.1 The effect of cables on a twoterminal component
With reference to Figure 5.19,
U
0
I
0
Z
c
2
I
0
I
Y
c
5:1
U
I
0
I
Y
c
IZ
c
2
5:2
But U/I Z, so (5.2) becomes
I
I
0
1
Z
c
Y
c
2
ZY
c
1
5:3
whence, from (5.1),
U
0
I
0
Z
c
2
Z
c
=2 Z
1 Z
c
Y
c
=2 ZY
c
5:4
Clearly, even if the YZ terms are negligible, there is a firstorder addition of Z
c
to Z. If we view a connector as being a cable between Z and a network, even
substitution measurement will be in error by the uncertain connector impedance Z
c
,
and if Z is a large impedance, the term Y
c
Z may be appreciable for even a short
length of cable or a goodquality connector. Twoterminal standards having coaxial
ports are only useful at the frequencies with which we are concerned in this book
when moderate accuracy is required over a limited range of values.
Paradoxically, they are common at microwave frequencies where the defined
conditions are those across a defined a plane of the conductor.
5.4.2 The effect of cables on a fourterminal coaxial component
A good definition of the arrangement of section 5.3.5 is obtained when connecting
a component of this kind to a network with coaxial cables only if one cable carries
the current to and fro from the component and the potential difference with zero
current flow is maintained between inner and outer conductors of the other cable.
Z
I
U U'
I'
I' I
Y
c
Z
c
/2 Z
c
/2
Figure 5.19 A cable added to a twoterminal component
96 Coaxial electrical circuits for interferencefree measurements
Representing the impedance Z with cables attached as shown in Figure 5.20,
from the lefthand mesh,
I I
1
I
0
Y
1
I I
1
Z
1
2
IZ 0 5:5
And from the righthand mesh,
I
1
Z
2
2
1
Y
2
IZ 0 5:6
Eliminating I
1
between (5.5) and (5.6),
I
0
I 1
ZY
2
ZZ
1
Y
1
Y
2
1 Y
2
Z
2
=2
Y
1
Z
1
2
Y
1
Z
From (3.8),
U
0
U
1 Y
2
Z
2
=2
Hence, the fourterminal admittance sensed at the end of the coaxial cables is
1
Z
0
I
0
U
0
1
Z
1
Y
1
Z
1
2
Y
2
Z
2
2
Y
1
Z Y
2
Z
Y
1
Y
2
ZZ
2
2
Y
1
Y
2
ZZ
1
2
Y
1
Y
2
Z
1
Z
2
4
5:7
Z U
Z
U'
I'
I
I
1
I
1
Y
2
Y
1
a
b
I' I I
1
I + I
1
I = 0
Z
1
/2 Z
1
/2 Z
2
/2 Z
2
/2
(b)
(a)
Figure 5.20 A (a) fourterminal component with cables attached (b) its equivalent
circuit
General principles of accurate impedance measurement 97
since 1/Z I/U. The product correction terms can readily be evaluated from the
properties of the cables and an approximate value of Z.
Typically, for a metre of coaxial cable of 50 Wcharacteristic impedance, where
Z is in ohms, Y is in ohms
1
and w in rad/s.
Z 10
2
jw2:5 10
7
Y jw10
10
(L
2
0.25 mH/m, C
2
100 pF/m, R
2
0.01 W/m)
The values of the termY
1
Z
1
/2, Y
2
Z
2
/2 are then w
2
2.5 10
17
jw2.5 10
12
.
At frequencies below w 10
5
rad/s (&16 kHz), these terms will not exceed
3 10
7
. The term Y
1
Y
2
Z
1
Z
2
/4 is usually quite negligible. These terms are inde
pendent of the value of the impedance Z being measured, but the remaining second
order terms Y
1
Z and Y
2
Z, and fourthorder terms are not. We consider three special
cases with cables having the properties listed above.
In the following numerical examples w is in rad/s, R is in ohms, C is in farads
and L is in henries.
(i) Z is a pure capacitance, C.
Y
1
Z
jw10
10
jwC
10
10
C
as expected, because C is merely shunted by the cable capacitance. Clearly,
the terminal configuration considered in this section is unsuitable for all but
large values of capacitance.
(ii) Z is a pure inductance, L.
Y
1
Z jw10
10
jwL w
2
10
10
L
This terminal configuration is not ruled out for moderate values of w and L.
For example, at w 10
4
rad/s (&1.6 kHz) and L 10
3
H, Y
1
Z 10
5
.
This is almost a negligible correction to a measurement of an inductance
because the accuracy is usually limited by both a lack of stability and, possi
bly, a large external magnetic field of the inductor, which can be affected by
its surroundings. The term Y
1
Y
2
ZZ
1
/2 jw10
20
(10
2
jw2.5 10
7
)L,
and for moderate values of w, this is smaller than the term just evaluated.
(iii) Z is a pure resistance, R.
Y
1
Z jw10
10
R
This is in quadrature with R, but at w 10
4
rad/s (&1.6 kHz), it becomes
appreciable for values of R that exceed 10
3
W if the cable shunt admittance
constituting Y
1
is a lossy capacitance. If this capacitance has a phase angle of
10
3
, this term will give an inphase addition to R of w10
10
10
3
R,
which, for w 10
4
rad/s, is 10
9
R. Therefore, this terminal arrangement is a
poor way of measuring R accurately at this or similar frequencies if the value
of R is greater than 1 kW.
98 Coaxial electrical circuits for interferencefree measurements
The fourthorder term Y
1
Y
2
ZZ
1
/2 is again much smaller for moderate
values of w than the term just evaluated.
5.4.3 The effect of cables on a twoterminalpair component
This case is shown in Figure 5.21, where the defining conditions are again to be
fulfilled at the ends of added cables instead of at the internal defining points. From
(3.7), we have I I
0
(1 Y
2
Z
2
/2), if Y
1
is small compared to 1/Z
2
. Since current flows
through the original terminals where the potential was defined, we cannot use (3.8) to
deduce the relationship between U and U
0
. We can carry out an approximate ana
lysis, however. If Y
1
and Y
2
are less than or of the order of 1/Z, then the current
through Y
H
is UY
H
and the current through Y
1
is approximately UY
1
. Therefore,
U
0
U
Z
1
UY
1
UY
H
I
2
Z
1
UY
H
I
2
or
U
0
U 1
Y
1
Z
1
2
Y
H
Z
1
Z
1
Z
Since Z U/I and the apparent impedance of the component and added cables
Z
0
U
0
/I
0
,
Z
0
Z 1
Y
1
Z
1
2
Y
H
Z
1
Z
1
Z
1
Y
2
Z
2
2
5:8
Hence, the correction to be made to the apparent impedance Z
0
to obtain Z
depends to first order on the term Z
1
/Z. This might have been anticipated since Z
1
is
in series with Z. The correction is subject to the uncertainty in the connector
impedance contained in Z
1
. Either satisfactory corrections can be made to account
for the other two terms, or the impedance can be regarded as having been redefined
with the lengths of additional cable included (but take note of section 3.2.6
regarding the nonlinearity of cable corrections if further lengths of cable are added
to include the component in a measuring network). Because of these considera
tions, a twoterminalpair definition is suitable only for highimpedances Z.
Since the way we define a threeterminal impedance is not different in prin
ciple from a twoterminalpair definition, the above considerations apply to this
definition also.
Z
Z
I'
U'
Z
1
/2 Z
1
/2
Y
1
Y
2
Y
H
Y
L
U
Z
2
/2 Z
2
/2
I
U = 0
Figure 5.21 Cables added to a twoterminalpair component
General principles of accurate impedance measurement 99
Many standard impedances made and sold for measurements of the highest
accuracy have twoterminalpair terminations, and no information is available
for the length or impedance and shunt admittance of the internal cables between
the impedance itself and these terminations. To improve their electrical definition,
Tconnectors can be added at the terminations. The impedance can then be regar
ded as being defined as a four terminal pair having its internal defining points at
these Tconnectors (see Figure 5.22).
The internal cables then become part of the defined impedance of the standard.
They alter its value, frequency dependence and phase angle, but the redefined
standard has fourterminalpair completeness.
One design of commercial capacitance standard has only one side of the capa
citor completely shielded. The other side is exposed to other circuitry through which
noise currents can flow and make the measurement unnecessarily difficult. Worse,
currents at the measurement frequency can alter the result by an unknown amount.
Because the exposed side is, in most measurement networks and instruments, con
nected to a source of lowoutput impedance through which most of the interfering
current is shunted away, the effect is often negligible. The exposed side is designated
Hi, and the other Lo. By taking care to connect the standard the correct way
round, reasonably reproducible and accurate results can be obtained, but it is clearly
better to modify these standards by completing the shielding. This involves dis
mantling the standard and adding a shield of a somewhat complicated shape.
5.4.4 The effect of cables on a fourterminalpair component
With reference to Figure 5.23, the addition of a cable between the generator and port 1
has no significance as no quantities are defined at this port. A cable added to port 4
also has no effect since if U0 and I 0 at one point on a cable, these conditions must
hold everywhere and, apart from sensitivity considerations, its length is immaterial.
We need to consider only the effect of adding cables to ports 2 and 3.
From (3.7), reversing the roles of I and I
0
, we have
I
0
I 1
Y
3
Z
3
2
and from (3.8),
U
0
U
1 Y
2
Z
2
=2
Z
T
H
T
L
P
L
P
H
C
H C
L
2TP component
Figure 5.22 Converting a twoterminalpair standard to four terminal pair by
adding Tconnectors at its output ports
100 Coaxial electrical circuits for interferencefree measurements
Hence, the apparent impedance Z
0
, after the addition of cables, is
Z
0
U
0
I
0
Z
1 Y
2
Z
2
=2 1 Y
3
Z
3
=2
5:9
Notice that the correction terms are now much simpler. It is especially
important also to note that they do not involve the impedance Z. This means that a
fourterminalpair definition is suitable for all values of Z, that is, for all com
ponents at all frequencies within our restriction that the wavelength is negligible
compared with the cable lengths. The effects of variable connector impedance,
which we can view as being contained in Z
2
and Z
3
, although still present, are the
same for all values of Z. By making a sensible choice of connector types and
restricting the values of Y
2
and Y
3
, the uncertainty introduced into the value of Z can
be kept less than 1 part in 10
9
at frequencies below 16 kHz.
For metre lengths of typical cable at 1.6 kHz, the correction terms are of the
order of 1 in 10
9
and are easily calculated from the measured or quoted properties
of the cables. The length is not very critical: at 1.6 kHz, a change of 100 mm in 1m
cables having the constant impedance and admittance per unit length assumed in
section 5.4.2 changes the fourterminalpair admittance by only 1 in 10
10
.
Sometimes Z is taken as having been redefined at the new ports at the ends of
the added cables, which then become part of the defined impedance Z
0
. Due regard
must again be paid to the nonlinear addition of the effects of cables external and
internal to the device, as pointed out in section 3.2.6.
There are three ways of configuring the internal cables of a fourterminalpair
impedance.
1. Two coaxial leads go from the impedance to an insulating panel on which four
coaxial connectors are mounted, and appropriate connections are made there
Z
I'
Z
U'
Z
1
/2 Z
1
/2
Z
2
/2 Z
4
/2
Z
4
/2 Z
2
/2
Z
3
/2 Z
3
/2
Y
L
Y
H
C
H
C
L
P
H
P
L
Y
1
Y
4
Y
2
Y
3
U = 0
I = 0 I = 0
Figure 5.23 Cables added to a fourterminalpair component
General principles of accurate impedance measurement 101
between inner conductors and between outer conductors to produce a four
terminalpair standard. The fourterminalpair definition can be strictly obeyed
at these connections and any corrections must be made for defining cables,
between them and a measuring network. The problem with this method of
cabling is that the two cables between the impedance and the panel modify the
defined impedance and, in particular, its frequency dependence and phase angle.
2. Four coaxial leads go from internal defining points of the impedance to four
ports mounted on an insulating panel. Making correct cable connections for
the defining cables, which connect this impedance to a measuring network,
now requires some care. The properties of the relevant cables between the
impedance and the panel have to be known and must be combined with those
of the defining leads outside the device as described in section 3.2.6, and then
the effect of the internal leads subtracted in order to produce a measurement as
if the fourterminalpair defining conditions had been strictly fulfilled at the
ports mounted on the panel. Alternatively, the defining conditions can be
regarded as holding at the internal defining points rather than at the panel
connectors.
3. With the configuration of 2 above, the defining conditions can be regarded as
holding at the potential ports on the panel. They can be strictly fulfilled by
detecting zero current through the highpotential defining port and connecting
a combining network (see section 5.6) to the lowcurrent defining port. This
approach has the disadvantage that the highpotential lead between the panel
and the ratio device (usually a transformer) loads this ratio device, and this
loading has to be accounted for. This can be done either by Tconnecting an
admittance having a known ratio to that of the loading cables and extrapolating
back to the condition of zero loading.
Although the complexity of networks for measuring fourterminalpair impe
dances is greater than that for the other definitions, the use of impedances defined
in this way is to be encouraged, and in work of the highest accuracy, it is essential.
The planes where a voltage difference or a current is defined can be moved along a
cable by employing triaxial cable with the intermediate screen driven by an
auxiliary source to a potential equal to that of the inner conductor, as explained in
section 3.2.7.
5.5 An analysis of conductorpair bridges to show how the effect
of shunt admittances can be eliminated
In this section, we examine how a bridge network can compare the direct admit
tances of devices whilst ignoring their input and output shunt admittances. (These
quantities have been defined in section 5.3.2.) It is possible to so arrange the
components that the bridge balance condition can be made to depend mostly on the
direct admittances but with a correction term that involves the shunt admittances.
102 Coaxial electrical circuits for interferencefree measurements
This term can be factorised into the product of two other terms, both of which
depend on shunt admittances but both of which can be made small by auxiliary
adjustments. The product is then usually negligible. We will discuss twoterminal
pair components for simplicity, as the shunt admittance aspect of bridge design is
not significantly different for fourterminalpair components. We will consider just
the three most commonly occurring cases, namely bridges based on a voltage
transformer ratio (see section 9.1.5), the classical generalised fourarm bridge dis
cussed in section 5.1 (which is still useful, particularly at higher frequencies) and
the network known as a quadrature bridge (see sections 9.2.19.2.3) because it
compares voltages across resistances with voltages in quadrature across capaci
tances. Some elements of the discussion can be recognised as the Wagner and
detector auxiliary balances of the older bridge techniques (see section 5.1) now
applied to the betterdefined conductorpair measurements.
Thompson [4] has also given an elegant treatment.
5.5.1 Comparing direct admittances using voltage sources
Consider the arrangement shown in Figure 5.24 where twoterminalpair com
ponents are connected to separate sources. Their direct admittances Y
1
and Y
2
are to be compared by adjusting either U
1
or U
2
or the ratio U
1
/U
2
to null
the detector.
I
I
D
I'
I'
U = 0
Y
2
Y
1
Y
U,1
Y
I,1
Y
U,2
Y
I,2
U
1
U
2
Figure 5.24 Comparing twoterminalpair components
If the sources are of negligible output impedance (e.g. if they are the outputs of
two separate windings of a wellconstructed voltage ratio transformer), their output
voltages U
1
and U
2
are not greatly affected by the presence of the shunt admit
tances Y
U,1
and Y
U,2
. That is, to a good approximation, Y
U,1
and Y
U,2
do not affect
the bridge balance. Also, when the detector is nulled, Y
I,1
and Y
I,2
have no voltage
across them. Hence, no current flows through these admittances, and their value
also does not affect the bridge balance condition, which depends only on the well
defined direct admittances Y
1
and Y
2
. The effect of Y
I,1
and Y
I,2
, which shunt the
detector terminals, is to degrade the detector sensitivity and noise performance.
The error arising from the impedances and shunt admittances of the cables
connecting Y
1
and Y
2
to the detector and to the voltage sources has already been
calculated in section 5.4.3.
General principles of accurate impedance measurement 103
5.6 Combining networks to eliminate the effect of unwanted
potential differences
5.6.1 The concept of a combining network
Historically, the concept of a combining network was first employed in Kelvins
double bridge. This network was invented to solve the problem of comparing
accurately the values of lowvalued fourterminal DC resistors. The fourterminal
defining conditions of these resistors are that the defined resistance is the ratio of
the voltage between the two designated potential terminals (under conditions that
no current flows through either) to the current leaving one designated current
terminal. The network is drawn in Figure 5.25.
D
R
R
1
R
3
R
4
R
2
r
2
r
1
I
a b
c
d
e
f h
g
Figure 5.25 A DC Kelvin double bridge. The components within the dotted box
are a combining network
The network is basically a Wheatstone bridge. When the detector is nulled by
adjusting R
4
, the ratio of the fourterminal resistances R
1
/R
2
approximately equals
the known ratio of the much highervalued resistances R
3
/R
4
. The potential term
inals g and h of R
1
and R
2
have to supply the small but finite current that traverses
the highervalued resistances R
3
and R
4
. Consequently, the defining conditions at
these terminals are only approximately fulfilled, but we are not concerned here with
this imperfection.
The current I traversing R
1
and R
2
causes a potential difference IR across the
resistance R, which is the total resistance between the low internal defining points
of R
1
and R
2
. This potential difference appears across the potential terminals a and
b between which the resistors r
1
and r
2
are connected, and the bridge detector is
connected to their junction point. A value of r
1
/r
2
can be found such that the
detector does not respond to changes in the potential between the low internal
defining points of R
1
and R
2
brought about, for example, by temporarily opening
the link between e and f. Then we can appeal to linear network theory to remark
that since the bridge balance no longer depends on the value of the potential dif
ference between the low internal defining points of R
1
and R
2
, the situation is the
same as if this potential difference were truly zero, the low internal defining points
104 Coaxial electrical circuits for interferencefree measurements
of R
1
and R
2
were actually connected together and the defining condition of zero
current through their lowpotential terminals a and b had actually been achieved.
The resistance of these potential leads between the internal defining points and a
and b forms part of the resistances r
1
and r
2
.
Evidently, r
1
and r
2
, together with the rest of the network, form an auxiliary
Wheatstones bridge balanced so that the detector does not respond to the source
IR. If the main balance condition R
1
/R
2
R
3
/R
4
has also been achieved, by itera
tion if necessary, their values are such that
r
1
r
2
R
1
R
2
R
3
R
4
5:10
That is, the voltage developed across the resistance between the low internal
defining points of R
1
and R
2
is divided by the resistances r
1
and r
2
at the point c in
the same ratio as the voltages between the internal defining points of R
1
and R
2
and
the bridge balance is independent of R and the network behaves as if the correct
defining condition of zero current through the lowpotential terminals of R
1
and R
2
has actually been achieved.
r
1
and r
2
form a combining network, which combines the two potentials at the
two internal defining points to which they are connected with the intention of
eliminating the effect of the difference in these potentials from the bridge balance
condition.
Warshawsky [5] has devised a network, which uses four combining networks
at the corners of a Wheatstone bridge, which allows fully correct fourterminal
definitions of all of R
1
, R
2
, R
3
and R
4
.
The technique of incorporating a combining network to eliminate the poten
tials between internal defining points of components has found widespread appli
cation to AC networks (bridges), which compare impedances accurately. There
are many examples later in this book.
5.6.2 A general purpose AC combining network and current source
If we consider the Tnetwork of Figure 5.26a, where nodes a and b are connected
by an ideal transformer of infinite input impedance and zero output impedance, the
current I flowing into a third node c is
I U
b
rU
a
U
b
U
c
Y
Consider the network of Figure 5.26b.
I
0
rU
a
U
c
Y 1 rU
b
U
c
Y I
for all values of r. Therefore, the two networks are equivalent.
Note that if a or b are taps on the transformer, values of r < 0 and r > 1 can be
obtained. Therefore, if r < 0, the admittance rY between a and c is negative, and
could be either a negative conductance or a negative susceptance. The admittance
between b and c would then be greater than Y.
General principles of accurate impedance measurement 105
If we take two such networks and connect them in parallel, as shown in
Figure 5.27a, the resulting network will be equivalent to that of Figure 5.27b, and
since r
1
and r
2
are fully adjustable, the network is a general combining network
having ratios of admittances of both in phase and in quadrature. It can therefore
eliminate the effect on the rest of the network of any potential difference between
a and b.
A sufficiently close approach to an ideal transformer can easily be made in
practice, and sufficiently large admittances G and jwC can be used so that there is
no significant effect on the noise (see section 2.1.3) and sensitivity of the total
network containing the combining network. We give several examples of its use in
the networks to be described in chapter 9. It is interesting that this AC combi
ning network is one of a dual pair of components, of which the other is an adjus
table quasicurrent source used extensively in the bridge networks described in
chapter 9.
Where two fourterminalpair impedances are connected in series via their C
L
terminals as shown in Figure 5.28 and the effect of the potential difference between
their internal defining points is eliminated with a combining network, the effect is
as if a and b were at the potential of the outer conductor when the detector D is
nulled.
Y
U
b
U
a
U
a
U
b
U
c
U
c
b
a
b
a
c
I
r
c
I'
rY
(a) (b)
(1r)
(1r)Y
Figure 5.26 (a) An AC combining network and (b) its equivalent circuit
jwC
b
U
a
U
b
a
c
c
G
r
2
j
w
C
r
1
G
b
r
2
r
1
a
(a) (b)
(
1
r2
)
j
w
C
(
1
r1
)
G
Figure 5.27 (a) Connecting two AC combining networks in parallel to obtain an
inphase and quadrature adjustment. (b) The equivalent circuit
106 Coaxial electrical circuits for interferencefree measurements
5.7 Connecting twoterminalpair impedances in parallel
While it is not in general easy to connect impedances in series and compute their
combined impedance because of the influence of the somewhat uncertain shunt
capacitances of the joining cables, connection of twoterminalpair impedances in
parallel is possible. For example, connection of two or more twoterminalpair
capacitances in parallel is a practicable way of producing a larger capacitance,
calculable from the individual twoterminalpair values as measured at the ends of
their cables. The junction points of these cables, when the capacitances are con
nected in parallel, then become the internal defining points of the parallel assembly,
which can therefore be defined as a fourterminalpair impedance if required. The
meshes formed by the parallel connection, in principle, need current equalisers.
References
1. Hague B. (revised by Foord T.R.). Alternating Current Bridge Methods.
6th edn. London: Pitman; 1971
2. Cutkosky R.D. Fourterminalpair networks as precision admittance and
impedance standards. Trans. IEEE Commun. Electron. 1964;83:1922
3. Kibble B.P. Four terminalpair to anything else! IEE Coll. On Interconnections
from DC to Microwaves (No. 1999/019) pp.6/16/6, 1999
4. Thompson A.M. AC bridge methods for the measurement of threeterminal
admittances. IEEE Trans. Instrum. Meas. 1964;IM13:18997
5. Warshawsky I. Multiple bridge circuits for measurement of small changes in
resistance. Rev. Sci. Instrum. 1955;26:71115
Z
1
Z
2
D
generator
a b
Figure 5.28 Connecting fourterminalpair impedances in series using a
combining network
General principles of accurate impedance measurement 107
Chapter 6
Impedance standards
This chapter details the key SI primary and secondary standards of capacitance, resis
tance and inductance for applications covering the frequency range DC to 100 MHz.
These standards include the ThompsonLampard calculable crosscapacitor, Gibbings
quadrifilar resistor, Campbell mutual inductance standard as well as the latest standard
of impedance derived from the quantumHall effect (QHE). Impedance measurements
span one of the widest ranges of any physical quantity, from nanoohms to teraohms
(i.e. a ratio of 10
21
). This chapter also discusses the limitations of the standards and their
evolution to meet the contemporary scientific and industrial needs.
Oliver Heaviside was the first to introduce the concept of electrical impedance
in the 1880s. Impedance is a more general concept than resistance because the latter
may be regarded as the value of a particular physical quantity (resistance) at a
specific point (i.e. at DC or f 0 Hz) on the frequency axis. In addition, impedance
takes phase differences into account for f > 0 Hz and is a fundamental concept in
electrical engineering. That is, it must be quantified by two parameters rather than
the one parameter of DC resistance. Thus, impedance may generically be expressed
in terms of complex numbers (or displayed on a polar or Argand diagram) such that
Z(w) R(w) jX(w), where w 2pf is the angular frequency, j 1
p
, R(w) is
the frequencydependent resistance and X(w) is the frequencydependent reactance.
The impedance Z(w), in general, can be any combination of resistance, inductance
or capacitance and can be measured by either energising a device with a known
voltage U(t) and measuring the current I(t) through it or vice versa, such that
Z(w) U(t)/I(t) Ze
jwt
. The real and imaginary components of the impedance
are then given by
ReZ Z j j cosf and ImZ jZj sinf 6:1
where
jZj R
2
w X
2
w
_
and f tan
1
Xw
Rw
_ _
6:2
There are four fundamental properties of impedance, as defined earlier and
discussed throughout this book; causality, the response of the system is due only to
the applied perturbation; linearity, the impedance is independent of the magnitude
of the perturbation provided the perturbation is not too great; stability, the system
returns to its original state after the perturbation is removed; finitevalued, the
impedance must be finitevalued as w ? 0 and w ? ? and must also be a con
tinuous and finitevalued function at all intermediate frequencies. As discussed in
section 6.4.2, these four conditions are implicit in the KramersKronig integral
relations (for an ideal system), which relate the real and imaginary parts of the
impedance spectra over a wide range of frequencies. In this chapter, we also briefly
discuss the empirical Debye and ColeCole relaxation models for gas/solid
dielectrics (or insulators with negligible conductivity), because any impedance
standard will inevitably be surrounded by, or will principally involve, a dielectric in
its physical construction. These complex theoretical developments are com
plemented by equally complex and elaborate array of measurement techniques,
which are discussed in more detail in the other chapters.
6.1 The history of impedance standards
Historically, the primary impedance standard was entirely arbitrary. In a laudable
attempt to include it in the number of standards based on naturally occurring phe
nomena, the ohm was defined as the DC resistance of a column of pure mercury
1 m long and 1 mm
2
in cross section. Wire resistance standards copied and dis
seminated this for practical everyday use. The accuracy was of the order of 1 in 10
3
.
Later, in the early 1900s, when AC impedance comparison bridges became
commonplace, the ohm was redefined in terms of a Campbell mutual inductor a
device whose impedance depends almost entirely on its geometry, which had to be
determined by making many dimensional measurements on the coils of wire it was
made from. The accuracy was much better of the order of 1 in 10
6
.
Later still, in 1956, the Australians Thompson and Lampard [1] devised the
calculable crosscapacitor that realised the SI ohm from the capacitance between
four long, parallel cylindrical electrodes. The improvement this standard was
achieved chiefly because the capacitance depends on the length of the capacitor so
that only a single length measurement is required, and this can be made very accu
rately with a laser interferometer. But a long chain of bridge measurements is needed
to relate the 2 10
8
W impedance of the capacitor at 1.592 kHz (w 10
4
rad/s) in
decade steps to the 11000 W values of maintained resistance standards. Even so, the
accuracy attained with different versions of this device in various national labora
tories is of the order of 5 parts in 10
8
. Another reason for preferring the calculable
capacitor to the older mutual inductor is that the inductor must be in the middle of a
considerable space free of conducting or magnetically permeable materials, whereas
the electric fields of a capacitor can be confined and completely defined by the
conducting walls of its enclosure.
Recently, attention has focussed on the quantised impedance of a QHE device.
At first, measurement chains begin with its DC resistance of about 12 906.4035 W.
This resistance is known in SI units because the von Klitzing constant, h/e
2
, which
governs this value, is closely related to the fine structure constant of atomic physics
[2]. The fine structure constant has been measured in SI units with an accuracy of
110 Coaxial electrical circuits for interferencefree measurements
0.68 parts in 10
9
by frequency observations made on the spectrum of atomic
hydrogen and by an independent method based on photon recoil from atoms.
A chain of measurements is needed to relate the 12 906.4035 W to the value of
practical working DC resistances, and another chain of AC measurements to relate
this value to that of working capacitance standards. It is important for the coherence
of the SI to discover whether the impedance unit set up in this way agrees with that
derived from a calculable capacitor because this would imply that the QHE is
indeed governed by h/e
2
to the accuracy of agreement obtained, with no unexpected
corrections to the theory of the QHE. At the moment, the best evidence for this is
that quantum Hall devices of various kinds including one based on the different
physical system of graphene have been shown to yield the same DC resistance
within a few parts in 10
10
. Figure 6.34 shows a quantum Hall device mounted for
DC or AC measurement.
6.2 The ThompsonLampard theorem
Dimensionally, capacitance is the product of a length, a shape factor and the permit
tivity of the space exposed to its electric field. Thompson and Lampards achievement
was to show how to avoid having to make any measurements to determine the shape
factor by finding a geometrical arrangement of the electrodes for which the capaci
tance is not dependent on the shape or size of the crosssectional dimensions perpen
dicular to the principal axis, provided that these crosssectional dimensions remain the
same as the principal axis is traversed over the measured length.
With reference to Figure 6.1, if the intersections of four electrode surfaces with
a plane perpendicular to their generators lie on the perimeter of a twodimensional
region of arbitrary and the shape of the region does not change, and if the gaps
C
2
C
1
Figure 6.1 The ThompsonLampard capacitance theorem
Impedance standards 111
between the electrodes are negligibly small, in principle, over an infinite distance in
the perpendicular direction, there is a relationship between the two capacitances per
unit length of an electrode and the nonadjacent one that is independent of the
shape of the region.
If the crosscapacitances are C
1
and C
2
between the two opposite pairs of
electrodes per unit length of the infinite system in a vacuum, this relationship is
exp
pC
1
e
0
_ _
exp
pC
2
e
0
_ _
1 6:3
where e
0
is the permittivity of vacuum. Practical ThompsonLampard capacitors
are invariably evacuated to avoid having to know the relative permittivity of air
between the electrodes. The relationship is not altered by the presence of thin,
longitudinally uniform layers of dielectric material on the electrode surfaces an
advantage of considerable practical importance since oxide layers and con
taminating coatings on polished metal surfaces cannot be entirely avoided [3]. We
note later that this property enables the device to constitute a phase angle standard.
Usually, the electrodes are made as symmetrical as possible so that C
1
C
2
DC,
where DC is small. Then (6.3) can be expanded as a series for the mean value
C (C
1
C
2
)/2.
C C
0
1
ln 2DC=C
0
2
8
_ _
% C
0
1 0:0866
DC
C
0
_ _
2
_ _
6:4
where C
0
is the value obtained when DC 0, that is,
C
0
e
0
ln 2
p
1:953 549 043 pF=m
assuming the defined value of m
0
4p 10
7
and the defined value of the
velocity of light c (e
0
/m
0
)
1/2
is 299 792 458 m/s. In practice, it is not difficult
to adjust the crosscapacitances C
1
and C
2
to be equal to better than 1 in 10
4
(i.e. DC/C
0
< 10
4
) so that (6.4) can be used to an accuracy of 1 in 10
9
.
Because circular cylinders are the easiest mechanical objects to make accu
rately, they are the electrodes in all existing ThompsonLampard capacitors. The
operative surfaces are the inwardfacing arcs shown in Figure 6.2b.
The need for electrodes of infinite length is overcome by inserting centrally up
the axis between the electrode surfaces, another conducting cylinder, as drawn in
Figure 6.2a. This movable electrode has a diameter that is slightly smaller than the
distance between opposite electrodes, and thus effectively blocks the cross
capacitances. The electrostatic field does not change along the axis of the electrode
systemexcept in the region that lies just beyond the end of this inserted cylinder. Here,
great distortions occur, but the distortion becomes negligible after a further distance
down the axis of two or three times the diameter of the inserted cylinder. Therefore,
since these end effects are the same with respect to the inserted cylinder no matter
112 Coaxial electrical circuits for interferencefree measurements
what its position with respect to the fixed electrodes, they can be eliminated by
inserting the cylinder a further distance l and observing the change in the cross
capacitances. l is of the order of 0.25 m, corresponding to a capacitance change of
about 0.5 pF, and is usually measured with a laser interferometer. The remote end
of the electrode system can be terminated by a similar fixed inserted cylinder as
shown in Figure 6.2a.
The main difficulty in the practical implementation of this beautiful concept
arises from the initial assumption that the electrode surfaces have straight generators
all parallel to an axis so that the cross section is always the same whatever the
distance along the axis. Imperfections in the manufacture of the electrodes will
inevitably mean that the electrode surfaces are not everywhere parallel to the axis,
and consequently the nonuniform electric field at the end of the inserted cylinder
will vary slightly with the position of this electrode, and the mechanical distance l
may no longer be quite identical with the electrical distance. It is usual to improve
this situation by providing a short extension to the cylinder of lesser diameter. For an
empirically determined ratio of length to diameter of this extension, the effect of a
slight linear divergence of the electrodes (e.g. if the electrodes are slightly inclined to
one another) can be compensated so that the mechanical and electrical distances can
again be made identical. The use of a properly adjusted extension is essential if
uncertainties of less than a part in a million are to be achieved with electrodes whose
departure from perfect straightness is of the order of a micrometre.
shield
electrode
movable
electrode
l
(a)
(b)
Figure 6.2 (a) A practical ThompsonLampard capacitor. (b) The inwardfacing
surfaces of four cylinders
Impedance standards 113
The operation of this extension can be understood by appreciating that dis
placing the four electrodes outwards increases the amount of electric field that
passes around the extension, and this compensates for the small extra amount of the
fringing field intercepted by the end of the extension. It is advantageous to provide
the fixed cylinder at the remote end with an identical extension because then the
crosscapacitances of the assembly will be immune to sideways displacements of
the electrodes from whatever cause (including thermal expansion).
The electrodes are not exactly equipotential surfaces because they possess
finite resistance and inductance. In particular, their inductance will cause errors
of the order of parts in 10
7
if it is not minimised, or a numerically calculated
correction made.
The type of bridge for relating the crosscapacitances to a fixed stable capa
citor is discussed in sections 9.1.4 and 9.1.5.
6.3 Primary standards of phase angle
Calibrations of electrical power measuring equipment, of power loss in reactive
components and of the phase angle of resistive components, need to be traceable to
a primary standard of phase angle. Fortunately the phase angle of a sealed two
terminalpair gas capacitor with clean electrode surfaces can be relied on to be less
than 10
5
. This is sufficiently accurate for all ordinary purposes, provided that the
electric field of the observed direct capacitance does not penetrate any lossy solid
dielectric used in the mechanical construction. There is nevertheless a need for a
primary standard of phase angle so that this statement can be made with con
fidence. There have been three main approaches to this problem.
In the past, mutual inductors at low or moderate frequencies have provided a
standard of phase angle. Departure from the ideal phase angle of the mutual reac
tance arises from eddy current losses in the conducting materials of the windings
and of any adjacent metallic material used for mechanical support or terminations
and from the capacitance of the windings combined with their resistance. The
energy associated with the electric field surrounding mutual inductors is small, but
it also causes loss in the dielectrics of insulation or the supporting structure.
Nevertheless, loss effects can be kept small and a phase angle standard with a phase
angle approaching 10
6
radians can be obtained. A mutual inductance phase angle
standard is particularly useful in calibrating wattmeters for the condition when the
current and voltage waveforms are in quadrature. The instrument should register
zero power under this condition.
A gasdielectric capacitor in which the spacing of the electrodes can be varied is
another approach [4]. A guard ring surrounds the lowpotential electrode to ensure
that no part of the electric field of the measured direct capacitance encounters solid
dielectric material. A correction can be made for the loss in the resistance of the
conductors connecting the electrodes to the terminals of the device. The only sig
nificant source of loss, which cannot be directly accounted for, arises from dielectric
films of water, metallic oxide, etc. on the surfaces of the electrodes. The phase defect
caused by this loss is inversely proportional to the electrode spacing and can be
114 Coaxial electrical circuits for interferencefree measurements
evaluated by varying it. Polished stainless steel is a suitable material for the elec
trodes, and the construction should minimise acoustic movement resulting from the
forces exerted on the electrodes by the electrostatic field. As these forces are pro
portional to the square of the applied voltage, this subtle cause of loss can be
investigated by observing the effect of varying the voltage applied to the capacitor.
The most precise primary phase angle standard makes use of a property of the
symmetrical ThompsonLampard crosscapacitor. If one of the electrodes is coated
uniformly with a lossy dielectric film, its contribution to the phase angle per unit
length of the two crosscapacitances is equal and opposite [3]. Neglecting small end
effects, it follows that the phase angle of the mean capacitance is independent of
axially uniform, but not necessarily identical, lossy films on the electrode surfaces.
The end effects can be eliminated by measuring the difference between two inserted
lengths of the fifth guard electrode, as in its usual role as a capacitance standard.
Shields [5] has adopted a slightly different approach by constructing a toroidal cross
capacitor that has no ends. The cancellation of phase angles is not theoretically
perfect as with the linear crosscapacitor, but is smaller than other causes of uncer
tainty such as acoustic loss, nonuniformity of any electrode surface film. A toroidal
device is also useful as a secondary maintained capacitance standard.
Rayner [6] has shown that the selfinductance and capacitance of an ordinary
electronic component resistor whose construction does not involve a cylindrical
spiral path, whose value R is of the order of 100 W and which is mounted in a very
loosely fitting coaxial tube, approximately cancel. The time constant t (L/R,
where L is the effective inductance) of this twoterminalpair device is less than
10
9
. Therefore, although it does not constitute a primary standard, this device
makes a very simple, practical standard whose phase angle is less than 10
5
at
frequencies below a few kHz.
Once a primary standard of phase angle has been established, the phase angle
of other standards can be calibrated from it by means of the bridges described in
sections 9.1.5 and 9.2.1.
A quantum Hall device in a proper measurement configuration (see section
6.7.5) probably constitutes an excellent phase standard, but this aspect has not yet
been fully investigated.
6.4 Impedance components in general
In this section, we describe the desirable attributes of resistance, capacitance and
inductance components, whether these are to be used as secondary standards in
impedance standardising laboratories or as circuit components in critical parts of
sensitive sensing and measurement circuits in general.
6.4.1 Capacitors
Capacitors have been exploited in a whole plethora of practical applications (such
as dielectric measurements, resonators, filters, energy storage, power conditioning
and sensing) ever since their inception over 250 years ago. Capacitance standards
Impedance standards 115
for impedance metrology are selected for their stability, small phase angle, small
frequency dependence and small change of value with applied voltage. The main
types of capacitance standards will be considered from the standpoint of their basic
physical and theoretical properties, which dictate their particular application. These
types are parallelplate capacitance standards (section 6.4.2), compact high
frequency transfer standards between high and low frequencies (section 6.5.8) and
a new type of capacitance standard that is coaxial and has a calculable frequency
dependence (section 6.5.9).
Electrical energy E is stored in a capacitor in the space between and immedi
ately adjacent to its two conducting electrodes. If the potentials of these electrodes,
as shown in Figure 6.3a, where the electric field is indicated by light broken lines,
differ by U, E U
2
/2C, where C is the capacitance. For example, for parallel plates
of area A separated by a distance L, which is small compared to the size of the
plates, C ee
0
A/L farads, where e
0
is the permittivity of free space (vacuum) and e
is the permittivity relative to this of the dielectric between the plates.
A
C
C'
R'
R
(a)
(b) (c)
U
L
e
0
e
1
Figure 6.3 (a) A parallelplate capacitor. (b) Parallel equivalent circuit.
(c) Series equivalent circuit
For a stable standard, it is clear that (i) a stable geometry of the conductors and
(ii) a constant dielectric relative permittivity are required.
Inevitably, some part of the electric field will spill out from the capacitor, in
the fringing field at the plate edges for example, or from unscreened terminals. Any
object in the vicinity of the capacitor that intercepts or modifies this field will alter
the measured capacitance. Hence, it is best practice to enclose the whole capacitor
and its terminals in a conducting screen. The effect of the screen on the capacitance
is large but definite and becomes part of the defined capacitance. Objects outside
the screen no longer affect the capacitance.
Two alternative lumped circuit representations of a lossy capacitor are drawn
in Figure 6.3b and 6.3c. The capacitive and resistive parameters C, C
0
and R, R
0
of
these equivalent representations are connected by the relationships
116 Coaxial electrical circuits for interferencefree measurements
C
0
C1 D
2
and R
0
RD
2
1 D
2
6:5
where
D
1
wCR
wC
0
R
0
tan d
1
Q
6.4.2 Parallelplate capacitance standard
Parallelplate capacitors have been employed to study the response of a variety
of materials (solid, liquid or gas) to both quasistatic and alternating electric
fields in order to ascertain their dielectric properties. The dimensionless real and
imaginary parts of the complex relative permittivity e e
0
e
r
e
0
je
00
of a
dielectric, where e
0
8.85 10
12
F/m is the permittivity of free space, contain
a great deal of information about the micro and macroscopic nature of the
material under investigation. The bulk of the work in this field, to date, has been
mostly concerned with dipolar, polar and ionic relaxation effects (where the
response of the induced polarisation in the dielectric is linear with the magnitude
of the applied stimulus voltage or current) up to around 10 GHz. For this pur
pose, the capacitor plates can be moderately large and have a variety of shapes
relevant to the particular application, such as rectangular, circular, coaxial or
interdigitated. Although the effect of the metallic electrodes on the measured
dielectric properties of the material is not usually a major concern, they can play a
significant role in electrolyte measurements, where interfacial polarisation effects
can dominate the measured response. Only in the last few years efforts have been
directed, at various research institutes, towards developing micro and nanoscale
capacitive, resistive and inductive devices and structures [7]. These make possible
investigations at frequencies as high as 1000 THz (i.e. optical frequencies) where
the dielectric properties can be dominated by electronic and atomic relaxation
effects. Clearly the subject of capacitive devices and their applications in material
measurements as well as in sensors and transducers is a very broad one. There
fore, the focus here will be confined to the use of parallelplate capacitors as
impedance standards and on their applications at frequencies ranging from power
frequencies (50100 Hz) to about 100 MHz, that is, the radiofrequency part of the
electromagnetic spectrum.
A capacitor can be a sensor responding to a number of physical conditions,
such as temperature, pressure, force, displacement, acceleration and humidity, but
in designing or selecting a secondary standard of capacitance these responses must
be minimised and, where necessary, specified in its calibration defining conditions.
Stability is important if the calibrated value of a capacitance standard needs to be
maintained as long as possible, in order to minimise the expense and time spent in
recalibration.
The original unscreened twoterminal connection configuration of a parallel
plate capacitor is still widespread, but, for accurate and reproducible measurements
Impedance standards 117
it is inadequate because of fringing field effects. This limitation became apparent
when demands on electrical metrology were increased after improved realisation of
the SI units. Moreover, commercial electronic instrumentation required more accu
rate calibration from national metrology institutes. Thus, a threeterminal definition
emerged (sometimes also referred to as guarded), which improved the reproduci
bility of the capacitance standards by introducing a shield conductor that surrounds
the main twoterminal capacitive element (see Figure 5.14). The direct capacitance
(see section 5.3.2) between the two active electrodes could now be measured highly
reproducibly in the presence of the shield conductor, with negligible influence from
external interference. Reproducibilty is gained if the extra shunt capacitance between
the high electrode and the shield merely loads the generator and that at the low
terminal has no potential across it. Therefore, neither shunt capacitance affects the
measured direct capacitance. It was not long before the limitations of the three
terminal definition and realisation also became apparent. The contact resistance
between the terminals of the standard and the connecting cables to the measurement
system (such as a bridge) lead to irreproducible changes in the measured dissipation
factor (or loss tangent) of the capacitance standard. Both the real and imaginary parts
of the dielectric constant must be determined for a complete measurement of a given
capacitor. To do this accurately, Cutkosky in 1964 [8] devised the fourterminalpair
definition and realisation of a capacitance standard. The beauty of this definition lies
in the universality of its application. The definition can be applied to capacitance,
resistance and inductance standards of all values. For formal two, three and four
terminalpair definitions of impedance standards, see sections 5.3.6, 5.3.7 and 5.3.8,
respectively.
As noted earlier, the electrostatic energy of a parallelplate capacitor is e CU
2
/2,
where C ee
0
A/d, the area of the electrodes being A and their separation d.
The permittivity e of the dielectric in the capacitor has real e
0
and imaginary e
00
components, and the latter represents an energy dissipation or real power loss P,
given by P wC
0
U
2
e
00
/2, or equivalently P we
0
e
00
AdE
2
/2, since E U/d and
C
0
e
0
A/d is the capacitance of a lossless capacitor of the same dimensions
having no dielectric. The impedance Z of a dielectricfilled capacitor may simply
be represented by a lumped electrical parameter model of a capacitor C in parallel
with a shunt resistor R
Z
1
G jwC
d
jwe
0
Ae
0
je
00
6:6
where G 1/R. Equation (6.6) is overly simplistic because in practice a real
capacitor will also have lead resistance, inductance, shunt capacitance and even
eddy current effects in the electrodes. These additional effects need only be
included in a more sophisticated model if greater accuracy is required for mean
ingful comparison with measurements. From (6.6), it is clear that the capacitance is
increased by a factor of e
0
over that of a vacuum gap (or air/gasfilled) capacitor.
The real and imaginary parts of Z are plotted in Figure 6.4 for a 1pF standard
118 Coaxial electrical circuits for interferencefree measurements
having an artificially low shunt resistance (for the purposes of this example only) of
1 GW at frequencies up to 100 kHz. The corresponding dissipation factor is then
given by D e
00
/e
0
1/wCR 0.159 rad at 1 kHz.
The graphs in Figure 6.4 show that the real part of the impedance is constant
at f < 100 Hz, whereas at higher frequencies it decreases quadratically with
frequency. In contrast, the imaginary part of the impedance initially increases
linearly with frequency and peaks around 200 Hz before eventually decreasing
linearly with frequency.
The real and imaginary parts of the permittivity can be calculated from the
parameters of a capacitor (measured with an impedance meter at, for example,
1 kHz) so that
e
0
Cd
e
0
A
and e
00
Gd
we
0
A
6:7
As shown in Figure 6.4, the real and imaginary parts of (6.6) are both fre
quency dependent, therefore the expressions in (6.7) may be regarded as low
frequency limits of (6.6). For example, for a 1pF capacitor with a more realistic
dissipation factor of 10
6
rad and a ratio of A/d 0.029 (typical for a fusedsilica
standard), it can readily be shown using (6.7) that e
0
3.8 and e
00
3.8 10
6
.
This compares well with data for fused silica from the literature of e
0
3.8 and
e
00
20 10
6
at 100 kHz and 20
C.
The frequency dependence of the real and imaginary parts (or equivalently
e
0
(w) and e
00
(w)) of the impedance given in (6.6) can, with care, also be established
from the Debye empirical model if three constants are known; the static or
2.0 10
8
0.0
4.0 10
8
6.0 10
8
8.0 10
8
1.0 10
9
frequency/Hz
10
2
10
3
10
4
10
5
10
1
R
e
(
Z
)
,
I
m
(
Z
)
/
Re(Z)
Im(Z)
Figure 6.4 Frequency dependence of the magnitude of real and imaginary
components of a 1 pF parallelplate capacitor at frequencies
up to 100 kHz
Impedance standards 119
lowfrequency (e
S
) and highfrequency (e
?
) permitivities and the characteristic
time constant (t) of the dielectric, such that
e
r
e
1
e
S
e
1
1 jwt
6:8
For a fusedsilica dielectric:
e
0
e
S
w
2
t
2
e
1
1 w
2
t
2
6:9
and
e
00
wte
S
e
1
1 w
2
t
2
6:10
whereas in the Debye model for an electrolyte capacitor (i.e. materials having a
small but finite conductivity, s e
00
w, as well as permittivity), the numerator in
(6.9) is replaced by (e
S
e
?
). The use of a single time constant can be justified
since the average response of the dipoles in the dielectric shows a resonant response
at a particular frequency (or relaxation time t CR). For completeness, (a) the
relative permittivity e
r
is related to the index of refraction n and the absorption
coefficient k (used at optical frequencies) through e
r
(n jk), such that e
0
n
2
k
2
and e
00
2nk, and (b) recently metamaterials have been developed that can have
negative relative permittivity as well as negative permeability topics that are
beyond the scope of this book. A good review can be found in Reference 9. In
practice, the Debye dispersion model is found not to be particularly accurate (since
it is based on empirical analysis and each dielectric material can have substantially
different features at the microscopic scale). As a result, ColeCole suggested a
slightly modified empirical model:
e
r
e
1
e
S
e
1
1 jwt
1 a
6:11
where 0 a 1 is a dimensionless fitting parameter that accounts for a distribu
tion of relaxation times, as found in real dielectric materials.
A fundamental theoretical model, which may be regarded as underpinning the
Debye and ColeCole empirical models for dielectrics, is the KramersKronig
transforms (which correspond mathematically to the Hilbert integral transforma
tion). The KramersKronig transformations are very general because they are
derived by assuming only the principle of causality (an effect cannot precede its
cause). They enable the imaginary part Z
00
(w) of an impedance to be calculated
from a knowledge for all values of the real part Z
0
(w) of the impedance for w
between 0 and infinity, or vice versa. It should be emphasised that none of these
models actually predict what the frequencydependent response of a given dielectric
will be. The former models are normally fitted to a set of determined measurement
data and relaxation constant, whereas the KramersKronig transformations enable
120 Coaxial electrical circuits for interferencefree measurements
calculation, usually numeric rather than analytic, of either the real or the imaginary
parts of the impedance, provided one or the other is known over a given frequency
range. Thus, the real part Z
0
(w) of the impedance is given by (6.12) in terms of the
imaginary part Z
00
(w) of the impedance:
Z
0
w Z
0
1
1
p
_
1
1
w
0
Z
00
w
0
wZ
00
w
w
0
2
w
2
dw
0
6:12
where Z
0
(?) is the highfrequency value of the real part of the impedance. Simi
larly, the imaginary part of the impedance is given by (6.13) in terms of the real
part of the impedance.
Z
00
w
w
p
_ _
_
1
1
Z
0
w
0
Z
0
w
w
0
2
w
2
dw
0
6:13
A difficulty that arises in evaluating these integrals is that a singularity (or a
pole) occurs when w w
0
. However, there are mathematical methods available
that circumvent this difficulty using the Cauchy principal integral. The key point to
note about (6.12) and (6.13) is that they enable an important crosscheck as to their
accuracy, validity and selfconsistency of any impedance spectra ranging from DC
to optical frequencies, or whether there are any systematic errors in the measured
data.
A further difficulty with KramersKronig transforms is the infinite frequency
range of measurements needed for accurate prediction of the conjugate impedance
spectra. Some mathematical methods combined with assumptions about the impe
dance spectra being finitevalued as f ??or f ?0 are usually used to circumvent
this problem. It can be shown that the Debye model can be derived using (6.12) and
(6.13). To date, it seems very little use of the above three models has actually been
made in the audio to radiofrequency range to complement impedance metrology.
This may be because these models are more applicable to electrochemical and
optical spectroscopy, but we see no reason why this should hinder their exploitation
in radiofrequency impedance metrology, albeit with care. To illustrate the appli
cation of the KramersKronig transforms, we apply them to a simple airdielectric
capacitor of C 1 nF, which has an overall series inductance of L 0.05 mH and a
dissipation factor of 10
4
rad at w 10
4
rad/s (or equivalently a series resistance of
R 10 W). This circuit will produce a series resonance at approximately 22.5 MHz,
given by
f
0
1
2p LC
p 6:14
The calculated real and imaginary parts of the impedance of the capacitor at
frequencies up to 1000 MHz are shown in Figure 6.5a as well as the corresponding
numeric evaluation of the KramersKronig (6.12) and (6.13). Figure 6.5b shows
the difference between the calculated and the KramersKronig relations (6.12)
Impedance standards 121
and (6.13). The agreement between the two methods is better than 0.5 mS (or 0.5%)
near the resonant frequency f
0
, but falls to 1.5 mS (or 1.5%) when the frequency of
interest is well below or above f
0
. Despite the differences at the higher and lower
frequencies, these results illustrate that comparisons with actual measured data will
give greater insight into the validity and accuracy of the assumed idealcomponent
inductancecapacitanceresistance (LCR) model and whether, for a given appli
cation, it requires modification.
6.4.3 Twoterminal capacitors
Capacitors in electronic circuits are usually unscreened twoterminal components.
Their conducting surfaces are often two foils separated by a thin solid dielectric
film, the sandwich then being rolled up tightly into a cylinder. Wires connected to
each foil are brought out at the ends. Another common construction is a stack of
alternating thin conducting and solid dielectric plates. Odd numbered conducting
plates are connected together to one terminal and the even to the other. Because
the plates or films are close together and the film between them has a high relative
permittivity, the fringing field outside the assembly is weak compared to that inside;
consequently, the capacitance associated with it is small, and neighbouring con
ducting or dielectric objects do not have any great influence on the apparent capa
citance. Nevertheless, these effects are sufficiently large to exclude these components
as accurate standards. For this purpose, they ought to be enclosed in a screen that is
usually connected to the high terminal to which a source is connected. Then currents
can flow from the screen to surrounding objects and return to the source without
forming part of the measured current flowing out of the low terminal.
6.4.4 Threeterminal capacitors
A much better reproducibility is achieved by connecting a twoterminal capacitor
to the inner conductors of high and low coaxial sockets and enclosing the
1
0.06
0.04
0.02
0.00
R
e
(
Y
)
a
n
d
I
m
(
Y
)
/
1
Im(Y)
ImKK(Y)
Im
0.02
0.04
0.06
0.08
(a)
0.10
0.12
10 100
f /MHz
Re
Re(Y)
ReKK(Y)
0.0020
0.0015
0.0010
0.0005
0.0000
0.0005
0.0010
0.0015
0.0020
Im
Y
K
K
(
Y
)
/
1
Im(Y KK(Y))
Re(Y KK(Y))
(b)
Re
1 10 100
f /MHz
Figure 6.5 (a) Calculated and estimated (from (6.12) and (6.13)) frequency
dependence of the real and imaginary components of a 1nF parallel
plate capacitor having 0.05 mH series inductance and D 10
4
rad at
frequencies up to 1000 MHz. (b) The corresponding difference
between the calculated and estimated values
122 Coaxial electrical circuits for interferencefree measurements
capacitor in an enveloping guard screen that is connected to the outer of the low
coaxial port. In the usual way of defining its measuring conditions, the guard is to
be maintained at the same potential as the low terminal, but not by being connected
to it. It should be connected elsewhere in the measuring network in such a way that
any current from it is diverted and does not form part of the measured current.
A configuration commonly employed in some gasdielectric parallelplate
standards is illustrated in Figure 6.6. The screen is connected to the outer of the
high coaxial connector, and a short insulated overlap limits current to the outer of
the low coaxial connector. This break in the screen is not readily apparent, and the
standard must not be mistaken for a twoterminalpair component.
C
C
Figure 6.6 A threeterminal capacitor and its diagrammatic representation
6.4.5 Two and fourterminalpair capacitors
The screen is connected to the outers of all the coaxial connectors. The added
complexity has the advantage, already explained in section 4.1.2, that the bridge
networks into which such a standard is connected can, by current equalisation, also
be made immune to external changing magnetic fields. The electrical definition of
these standards (see sections 5.3.6 and 5.3.8) is sufficiently precise for even the
most accurate measurements.
6.4.6 The mechanical construction and properties of various types
of capacitors
The stability and phase of capacitors under environmental influences such as
ambient temperature and humidity depend on their mechanical construction and
whether they have a solid dielectric or whether the plates are supported so that they
are separated and immersed in a dry gas or a vacuum in a sealed container. The
latter construction yields lowloss capacitors of 10 nF or less whose vacuum or gas
dielectric has a relative permeability that is independent of frequency.
For values greater than this, the large plate areas and small separations required
make construction impossible and so the electrodes must be separated by a solid
dielectric. Some capacitors of value less than 1 nF are made that have pure fused
silica as the dielectric, which have a phase angle of only 1 10
6
and which are
exceedingly stable, but unfortunately the fragile nature of this dielectric makes it
impracticable for constructing larger valued capacitors.
The principal types of highquality capacitors that can be used for standards
are listed in Table 6.1, which lists as a guide only the properties of various types of
Impedance standards 123
capacitors that might be used as standards or in other critical applications. The units
are partspermillion for the changes indicated in the table headings. The properties
of switched decade capacitance boxes are those of the capacitors contained in them,
with some increase in phase angle at the higher settings of capacitance because of
the resistance of the switch contacts and internal wiring.
The variation of the apparent value of a capacitor with frequency principally
arises from two causes. An example of the first cause would be the finite induc
tance of the connections to the electrodes of largevalued capacitors. The finite
resistance of these leads increases the apparent phase angle of the capacitor. A lead
geometry having orthogonal current and potential connections, which minimises
these effects, is described in section 6.5.7. The second cause is the frequency
dependence of the relative permittivity of any dielectric between the electrodes.
This is related to the loss in the dielectric, and a rule that is often applicable is that
the order of magnitude of the proportional decrease in capacitance on increasing
the applied frequency ten times is the same as the phase angle at the higher
frequency.
The electrode structure of vacuum or gasfilled capacitors can deform slightly
under their own weight. This distortion will depend on their orientation so that the
capacitance will change as the capacitor is tilted. In extreme cases for calibrations
of the highest accuracy, it may be necessary to specify the orientation at which the
capacitor is calibrated.
Electrode assemblies may also be prone to disturbance by vibration or acci
dental jarring through handling, and changes in capacitance may result. This pro
blem can be minimised either by making the electrode structure more rigid or by
ensuring that the capacitance as a function of electrode displacement in particular
directions is a maximum or a minimum and is independent of small electrode
displacements in others. A satisfactory design can result from a combination of
these techniques, and will have the additional advantage that the capacitance
change resulting from differential thermal expansion between parts of the electrode
Table 6.1 Types of capacitors and their properties
Type and
dielectric
Range of
values
Temperature
coefficient
per K
10% change
in relative
humidity
at 20
C
Barometric
pressure
effects
per kPa
Stability
per
year
Phase
angle at
1 kHz,
mrad
Sealed;
dry gas
01 nF 125 _ _ 15 15
Unsealed; air 010 nF 125 20 6 15 125
Fused silica 0100 pF 10 _ _ 1 15
Mica 1 nF1 mF 2050 525
(unsealed)
_ 525 50200
NPO or
COG
ceramic
1 pF0.1 mF 10 <1 0.05 1 100
Polystyrene 1 pF1 mF 200 _ _ 50 60
124 Coaxial electrical circuits for interferencefree measurements
structure will also be minimised. The temperature coefficient of capacitance then
arises from thermal expansion of the electrode structure as a whole. Thermal
expansion can be reduced by manufacturing from low thermal expansion materials
such as Invar or fused silica. Another approach is to design the geometry of the
electrodes so that thermal expansion of one part produces an equal and opposite
effect on the capacitance to the expansion of another part. These thermally com
pensated capacitors are unfortunately sensitive to transitory temperature changes
that produce temperature gradients within the electrode structure, and a period of
several hours at a constant temperature may be required for the whole capacitor to
be at the same temperature and the capacitance to become constant.
The best standards for work of the highest accuracy have a builtin resistance
thermometer; it is then possible to use as their temperature coefficient the ratio of
their change of capacitance to the change in resistance of the thermometer and to
refer their defined capacitance to a certain value of this resistance. This simplifi
cation means that it is never necessary to consider their temperature per se, but does
have the slight disadvantage that any drift in the resistance of the thermometer
appears as if it were a drift in the capacitance itself.
The AndeenHagerling company makes a range of instruments for measuring
twoterminalpair capacitances in the range 01 mF with an accuracy of 3 ppm, a
resolution of up to 0.5 aF, and accompanying loss tangents down to 10
8
. This
remarkable performance is achieved by basing these instruments on an internal
temperaturecontrolled parallelplate capacitance standards having fused quartz as
their dielectric and comparing this with the capacitor to be measured via a voltage
ratio transformer. Models are available that can measure over a frequency range
from 20 Hz to 20 kHz.
They also supply temperaturecontrolled capacitors in values of 1, 10 and
100 pF, housed in separate enclosures to serve as laboratory standards for general
calibration and intercomparison purposes. Other values can be made to special
order. Their remarkable properties include a relative capacitance stability of the order
of 10
7
/year, an exceedingly small hysteresis of capacitance when temperature
cycled, for example, between their 55
C controlled temperature and ambient tem
perature if the temperature control is switched off. They also transport well between
laboratories when relative changes of as little as 2 in 10
8
are found.
There is one characteristic to be aware of. As supplied, the internal wiring
between the inner conductor of the Hi coaxial port and the internal capacitor has
no screen between it and the temperaturecontrol circuitry. That is, the capacitors
have a threeterminal configuration with exposed Hi connections. This can lead to
measurement error and serious interference problems unless the Hi port is driven
from a voltage source of small internal impedance such as a ratio winding of a
voltage ratio transformer or the Hi connection of an AndeenHagerling bridge.
Even if this is done, there can still be a small measurement error.
Some laboratories have therefore dismantled their AndeenHagerling capa
citance standards and added shielding to the Hi side to convert the capacitors to a
twoterminalpair configuration. For national and international comparisons of
the greatest possible accuracy, the standards can easily be further modified to a
Impedance standards 125
fourterminalpair configuration by adding Tconnectors at their output ports (see
section 5.3.8). This modification has the advantage that no account need be taken
of the unknown cable properties within its enclosure when calculating cable
corrections these properties simply become part of the defined value of the standard.
Two or fourterminalpair capacitors of very small value, known as Zickner
capacitors, can be made by interposing between two plates a screen having a small
aperture in it. This screen is part of the surrounding conducting case of the capa
citor as shown in Figure 6.7.
Figure 6.7 A Zickner capacitor
By interposing interchangeable screens with apertures of different sizes any
capacitance from 1 pF to 0 (a screen with no aperture) can be obtained.
6.4.7 Capacitance standards of greater than 1 mF
The electronics industry uses large numbers of electrolytic capacitors and similar
types of large storage capacity. Values are typically from 1 mF to 1 F at frequencies
from DC to a few hundred hertz. Quality control and standardisation require that
the capacitance of these components be measured to a modest accuracy of the order
of 1%. To verify the accuracy of the commercial bridges that make these mea
surements, stable standards of at least the decade values within this range are
needed. Electrolytic capacitors themselves are not stable enough as a function of
time, applied voltage (AC or DC) or temperature, to be useful as standards. Non
electrolytic capacitors such as polystyrene film metal foilrolled capacitors are
good enough but are only available as single components in values up to 1 mF.
Other plastic film materials such as polycarbonates enable rolled capacitors to
be made in values up to 10 mF, and they can be connected in parallel to provide
standards of up to, say, 100 mF in value. Although their stability and phase angle are
not quite as good as polystyrenerolled capacitors, they are good enough for
the present purpose.
Standards of apparently greater capacitance can be made as a transformer
aided component. An elegant arrangement with two autotransformers tapped at
fractions a and b is shown in Figure 6.8a.
The direct admittance Y
T
of this device is the ratio of the current I flowing into
and out of one pair of terminals to the opencircuit voltage U that appears as a result
at the other pair. The device is therefore a fourterminal component having separate
126 Coaxial electrical circuits for interferencefree measurements
pairs of current and potential terminals. Y
T
is principally the admittance of a large
equivalent capacitance C
T
.
The transformers can be represented as equivalent Tnetworks, as shown in
Figure 6.8b, where Z
1
and Z
7
are small impedances arising mainly from the resis
tance of the transformer windings with a small contribution from the transformer
leakage inductances. Z
2, 3, 5 and 6
are primarily composed of the transformer winding
inductances that are designed to be much larger than 1/w
2
C.
Y
T
jwC1 1=w
2
LC
ab
where jwL is approximately the paralleled impedance of the two transformers:
jwL
1
% Z
2
Z
3
1
Z
5
Z
6
1
If we define C
0
C[1 (1/w
2
LC)],
Y
T
jwC
0
ab
and we have the very simple result that
C
T
C
0
ab
6:15
C
a
b
C
H
C
H
Z
1
Z
2
Z
6
Z
3
Z
5
Z
7
C
L
C
L
P
H
P
H
P
L
P
L
U, I = 0
U, I = 0
A
B
I
C
A
B
I
(a)
(b)
Figure 6.8 (a) Diagrammatic representation of a transformerenhanced
capacitance. (b) Its equivalent circuit
Impedance standards 127
C
0
is the apparent twoterminal capacitance between A and B with the trans
formers connected and the current and potential terminals opencircuited. C
0
varies
slightly with the voltage U because L depends on U, which affects the magnetic
flux density in the transformer cores. Provided that the voltage between A and B is
measured and a small predetermined correction made, this is not a serious limita
tion at the accuracies envisaged.
The principal disadvantage of this otherwise elegant idea is the impedance
present in the upper current and potential leads. These impedances can be expres
sed as
Z
C
Z
Co
aa b
jwC
0
6:16
and
Z
P
Z
Po
bb a
jwC
0
6:17
where Z
Co
and Z
Po
are the impedances measured in the C
H
and P
H
leads with A
shorted to B. The first term of each expression is the resistance and leakage
inductance of the transformer, and the second is a capacitance that is positive or
negative, or zero if a b.
To obtain a ratio between C and C
T
that is an odd power of 10 with this
restriction requires 1/a 1/b 10
n
H10. Since H10 is irrational, this cannot be
accomplished with a finite number of turns on the transformers, but a close
approximation can readily be obtained.
The General Radio model number 1417 standard is based on this principle. To
provide all the decade values from 1 mF to 1 F in one unit, the transformers have
switchselected tapping points. The transformers are wound with a large number of
turns of fine wire to ensure that the difference between C and C
0
is not too great
because the impedance of the inductance L of the transformers in parallel is made
much greater than that of the capacitance C so that variations in L do not cause
unacceptably large variations in C
T
. Unfortunately, as a consequence, the windings
have large resistances that result in somewhat large, mostly resistive, impedances in
the C
H
and P
H
leads for values other than 1 mF. These impedances are of the order
of the direct impedance for the 10 mF range at 100 Hz and rise to about 50 times the
direct impedance for the 1 F range at 100 Hz. Note that the impedances in the C
L
and P
L
leads are much lower as they arise from only a short length of connecting
wire, and this enables a standard of this kind to be compared with another using a
comparatively simple bridge such as that in section 9.1.3 and the fourterminal to
fourterminalpair adaptor in section 5.3.10.
The actual construction of the device incorporates networks that are associated
with the switching of the transformer taps and that provide a constant small phase
angle for the complete device that is independent of the value selected. By other
switched networks, the same phase angle is achieved at the three frequencies of
100 Hz, 120 Hz and 1 kHz.
128 Coaxial electrical circuits for interferencefree measurements
As an alternative design, if C is a 100mF capacitance made from poly
carbonate components, the number of turns needed on the transformers can be
greatly reduced. For example, the transformer ratios a and b could both be 127/400
turns for a 1mF value and 40/400 turns for a 10mF value. By having a and b equal
the second terms in (6.16) and (6.17) above are zero, and since C is 100 times the
value of that in the General Radio design, 1/10th of the turns are needed on the
transformers to give L a suitably high value. Therefore, they can be wound with a
heavier gauge of wire with a resulting lower value for the C
H
and P
H
lead resis
tances. The approximation of 400/127 to H10 results in a capacitance that is
slightly different from 1 mF, but it can be trimmed if required by shunting C with a
small shimming capacitance.
6.4.8 Voltage dependence of capacitors
Even welldesigned and constructed standard capacitors are not quite perfectly
linear circuit elements; their apparent value depends to some small extent on
the voltage applied to them. Mechanical movement of the electrodes under the
force exerted by the electric field between them and dielectric effects arising from
the strong field gradients at the edges of thinfilm electrodes coated onto solid
dielectric separators are among the causes of this phenomenon. For the best stan
dard capacitors whose dielectric is vacuum, dry gas or solid pure fused silica and
whose values are in the range from 0 to 1000 pF, changes are much less than a part
in a million for the applied voltages in the range from 1 to 100 volts, which are
commonly used in highaccuracy applications. Nevertheless, in the most accurate
work, the change with voltage must be accounted for, in particularly, when cali
brating voltage transformer ratios by the permuting capacitors method (see section
7.4.6). The value of a standard should be defined as that measured at a specified
applied voltage. This value might be the value obtained by extrapolation to a
negligibly small voltage. It is not possible to make a reference capacitance standard
having a guaranteed known small change of value with applied voltage, but the
change in value of a capacitor can be determined in terms of the increments pro
duced by factors of two increases in applied voltage, provided that three capacitors
of the same nominal value are available, together with 2:1 and 1:1 ratio trans
former bridges. The method of Shields [10], which is fully described in Kibble and
Rayner [11], assumes that the ratio of a properly constructed 2:1 voltage trans
former (Figure 7.46) can be measured to a sufficient accuracy, say 1 in 10
9
, either
by the method of section 7.4.3 or by exchange of nominally equal capacitors in to
1:1 voltage bridges centred on the 0 and 1 taps.
Recent improvements in making AC measurements of capacitors directly in
terms of the QHE [12] offer an alternative. The voltage coefficient of quantum Hall
devices can be adjusted to be very small by comparing them with resistors whose
voltage coefficient is negligible at the very low power applied to resistor and
device. Then the variation in the values of capacitors with applied voltage can be
obtained by comparison with quantum Hall devices at even lower values of power
dissipation.
Impedance standards 129
6.4.9 Resistors
The primary requirement for a resistor is that its resistance should be as constant as
possible with time and for the range of currents and frequency and with ambient
changes of temperature, pressure humidity, etc. within its design limits. For high
accuracy impedance measurements, the electric and magnetic fields that inevitably
accompany current flow in the device should be minimised so that the reactive part
of its impedance is small. Any energy loss associated with these fields should also
be small so that as much as possible of the energy dissipation takes place in the
resistance element. Undesirable losses include dielectric loss or loss arising from
eddy currents induced in either the surrounding metal such as other parts of the
resistance element (the proximity effect) and each individual part of the resistance
element (the skin effect). These effects are frequency dependent and therefore lead
to a frequency dependence of the resistance.
The principle cause of change with time is change in the state of strain of the
wire. All resistors are to a certain extent also strain gauges. The resistance element
should be mounted with as little mechanical restraint as possible, consistent with a
sound mechanical construction, and joints or connections between the resistive
material and other conductors must be made by a stable technique. Any jointing
technique such as softsoldering, crimping, pressurewelding and brazing will
produce a surrounding region of strain. The extent of this region should be
minimised.
Only chemically inert substances should be used in the construction. The
resistance element is often immersed or embedded in oil or a flexible potting
medium to exclude the corroding effects of the atmosphere and aid heat transfer.
The most successful material for the resistance element is a fourcomponent or
quaternary alloy of 75% Ni, 20% Cr, 2.5% Al, 2.5% Cu made under the brand name
of Evanohm. Its change of resistivity with temperature is of the order of ppm/
C
and is parabolic. The parameters of the parabola can be modified to alter the
resistance and its temperature coefficient to a limited but useful extent by the heat
treatment given to the wire as part of the strainrelieving process during manu
facture of the resistor. By positioning the maximum of the parabola near the desired
working temperature, the temperature coefficient at that temperature can be made
very small. Evanohm has a small thermoelectric Peltier coefficient with respect to
copper, which is important for DC and very low frequency applications where the
combination of Peltier or Thompson effects can alter the apparent resistance. Pel
tier heating or cooling at alloycopper junctions can combine with the thermal
capacity of the junction to produce a thermoelectric emf that can persist into the
next cycle of current reversal.
The main constructional techniques that attempt to put these principles into
practice are described in the following paragraphs.
The most stable resistors having values in the range from 0.1 to 10
4
W are
constructed with minimally supported bare wires. They are the resistance standards
of national and other calibration laboratories. The resistance wire is stout enough
(0.13 mm is usual) to be selfsupporting without touching neighbouring turns
130 Coaxial electrical circuits for interferencefree measurements
between the supporting points on the framework. The aim is to ensure that the wire
is not under tension so that no change in strain, and therefore in resistance value,
occurs. The winding on its supporting framework should enclose only a small open
loop area to minimise its inductance; less attention need be paid to stray capaci
tance as it is less important for these lower resistances. These Sclass resistors are
now designed to be suitable for AC impedance measurements, although DC
methods are still the principal means of intercomparison. Bindingpost terminals
provide a fourterminal definition that ought to be modified to a fourterminalpair
definition by connections to added coaxial sockets or by the adaptor described in
section 5.3.10 if accurate AC measurements are to be made.
The resistance element is sealed in an oilfilled metal container and the term
inals are fixed in an insulating top, or through insulating bushes. A well for a
thermometer lies alongside the resistance element. The entire resistor can be
immersed in a temperaturecontrolled oilbath or metal block.
Figure 6.9 shows the construction of a premier class standard resistor of this
type. The resistance element is supported on a skeletal insulating framework.
Somewhat loose winding and annealing at about 180
C for several days ensures
the vital strainfree condition.
Figure 6.9 An Sclass standard resistor
Standard resistors in the range from 10
3
to 10
6
W are made by winding enamel
insulated wire on to small thin cards, usually of mica. The reactance of these
resistors, particularly at the higher end of the range, will arise principally from
interturn capacitance. The dielectric loss associated with capacitances should be
small enough to ensure that the resistance is not altered significantly. The induc
tance of the wound resistor should also be kept small, but the fine wire, the large
Impedance standards 131
number of turns and their close proximity set a limit to what can be achieved.
Careful annealing helps ensure the stability of the resistance.
A range of resistors from 1 W to nearly 2 MW manufactured by Vishay are
relatively inexpensive and have properties that make them suitable as secondary
standards or components in networks where very good stability is required. Their
meandering ribbon thinfilm construction on a few millimetres square substrate of a
material that is closely matched to the temperature coefficient of expansion of the
resistive material confers low inductance and selfcapacitance so that they are
useful for both AC and DC applications. Their temperature coefficient is typically
2 ppm/
2
10
4
H
The actual residual inductance is unlikely to exceed 10
5
H so that the shunt
capacitance representation is appropriate. On the other hand, for a 1W resistor, the
equivalent inductance for a shunt capacitance of 1 pF is
L 10
12
1
2
10
12
H
and the actual residual inductance will certainly exceed this value; hence, the series
inductance representation is now likely to be appropriate.
The capacitive and inductive contributions can cancel, resulting in a very small
time constant. This is likely for a resistance value of the order of 100 W, the exact value
depending on the method of construction. This property is made use of in section 6.3.
Auxiliary components can be added to resistors to reduce their phase angle.
The two cases are treated separately.
(i) If the resistor has a predominately inductive residual reactance, it can be shunted
by a capacitance C, which, if wL ( R, can be adjusted to have the value
L/(R
2
w
2
L
2
) &L/R
2
. Then the total impedance is Z &R (see Figure 6.13).
A more elegant solution that compensates over a wide range of frequencies
is shown in Figure 6.14. An auxiliary resistor whose value is approximately
equal to R is connected in series with C. If C is adjusted to equal L/R
2
, Z R
for all frequencies for which the lumped parameter representation is valid.
(ii) The resistor has a predominantly capacitive residual reactance.
C
R L
R
Figure 6.14 Compensating an inductive resistor over a wide frequency range
C
R L
U
I
I
Figure 6.13 Compensating an inductive resistor
134 Coaxial electrical circuits for interferencefree measurements
The roles of inductor and capacitor in case (i) cannot readily be reversed for
this case. Whereas an adjustable capacitor of high impedance and small phase angle
can readily be obtained, this is not the case for an inductor.
A practicable solution is to design the resistor so that it has two nominally
identical sections with a tapping point between them, each section having a capa
citive phase defect. Phase angle compensation can then be carried out by con
necting an adjustable capacitance C
0
from the tapping point to the outer conductor,
as shown in Figure 6.15. A TD transformation of the network then enables it to be
represented as in Figure 6.16 and the direct admittance to be calculated as
Z
U
I
2R
2 jwCR
jwC
0
R
2
2 jwCR
2
For resistors of small phase angle, that is, if wCR (1 and C
0
2C, Z will be
real and equal to R.
U
I
C
R/2 R/2
C
I
C'
Figure 6.15 Compensating a capacitive resistor
Z
U
I
I
Figure 6.16 The equivalent direct impedance
In practice, the two sections will not quite be identical but C
0
can nevertheless
be adjusted to make the phase angle of the direct impedance of the device zero.
6.4.12 Mutual inductors: Campbells calculable mutual
inductance standard
Before the advent of the ThompsonLampard calculable capacitor in 1958 [1],
various NMIs employed either self or mutual inductance standards in order to
realise the unit of resistance and reactance. Selfinductance standards were rela
tively straightforward to construct and use in practice, because the value of resis
tance standards could be derived from their calculated values using bridge
comparison techniques. The accuracy of the calculated values for such standards
was severely limited due to the number of dimensional measurements that were
Impedance standards 135
needed. In addition, the mechanical instability of the standards also limited their
accuracy. Consequently, Campbell in 1907 [13] proposed a calculable mutual
inductance standard based on a coaxial set of coils.
A current I in a primary coil produces a proportional magnetic flux f that
threads an adjacent secondary coil and induces a voltage U. That is, f MI, where
the quantity M is called the mutual inductance between the primary and secondary
coils, and is in quadrature with I, since if I I
0
sin wt,
U
df
dt
I
0
wM cos wt I
0
wM sinwt 90
Mutual inductance has attributes that, in the past, led to mutual inductors being
widely used in bridges for reactance measurements. It is an additive quantity in that
if a primary coil induces magnetic fluxes in two or more secondary coils, their
individual induced voltages may be added. Therefore, neglecting capacitive effects,
a precise linear scale of mutual inductance can be obtained from a primary coil and
tapped secondary coils.
The primary winding of a Campbell mutual inductance standard consisted
of two singlelayer coils connected in series, wound on the same former, but
separated along the former by a distance equal to the radius of the coils, as shown in
Figure 6.17. The secondary of the mutual inductance standard was wound centrally
in between the two primary coils and consisted of several layers. This design
conferred several advantages. In particular, the calculated mutual inductance only
required five accurate dimensional measurements (such as the mean primary coil
diameter and pitch length, and secondary coil diameter, width and depth). In addi
tion, at a certain value of the radius of the secondary winding, the mutual inductance
almost becomes independent of small changes of this radius or axial displacement
(i.e. dM/da 0, where M is the mutual inductance and a the radius of the secondary
winding). In this configuration, the secondary coil is positioned such that all round its
mean circumference of the field due to the primary coils cancel. Campbell built and
A/2
A/2
A
a = 1.46A
Figure 6.17 A Campbell mutual inductor
136 Coaxial electrical circuits for interferencefree measurements
tested a nominal 10 mH mutual inductance standard that had 75 turns each for the
two primary windings wound on a marble cylinder of 150mm radius and a sec
ondary winding with 488 turns wound in a channel of 100 mm
2
with a mean radius of
219 mm (at a centretocentre distance between the primary and secondary windings
of b 150 mm).
The mutual inductance of the standard can be expressed analytically using
elliptic integrals. Thus, if A is the radius of the primary winding with n
1
turns and a
is the radius of the secondary winding with n
2
turns, then the mutual inductance is,
in SI units, given by
M m
0
n
1
n
2
A a
c
k
Fk Ek
A a
b
y
_ _
6:20
where m
0
4p 10
7
H/m is the freespace permeability, E(k) and F(k) are the
complete elliptic integrals of the first and second kind to modulus k, respectively, b
is the distance between the primary and secondary coil centres with
k
2
4aA
A a
2
b
2
sin
2
g c
2
4aA
A a
2
6:21
and
y
p
2
Fk Ek
0
; b Fk EkFk
0
; b 6:22
where E(k
0
, b) and F(k
0
, b) are the incomplete elliptic integrals of the first and
second kind, respectively, k
0
cosg, c
0
1 c
2
p
and b sin
1
c
0
=k
0
. Thus
(6.20) shows that the mutual inductance is calculable using only m
0
and linear
dimensions (hence, the emergence of the ThompsonLampard calculable capacitor
that similarly requires e
0
and, crucially, measurement of only one linear dimension,
see section 6.2). In practice, small corrections are made to the ideal expression for
mutual inductance given by (6.20) for the finite relative permeability of the former
and the nonuniform distribution of the current in the wires. Consequently, the
initial accuracy of the standards was limited to around a few parts in 10
5
, but was
later improved in the 1960s by, for example, G.H. Rayner at NPL, to around 1 part
in 10
6
.
Measurements of practical mutual inductance standards, for precision work,
were typically performed at low frequencies between 10 and 125 Hz using a bridge
comparison method because (6.20) is accurate only as the frequency f, at which the
inductor is operated, tends to zero. Therefore, operation at a higher frequency than
this also requires a small but finite allowance for the frequency dependence of the
mutual inductance. In a mutual inductance bridge measurement system, for the
ideal inductor the potential of the primary and secondary windings is made definite
by joining one end of both windings together as shown in Reference 11 p. 82.
Impedance standards 137
Consequently, the equivalent circuit of such a mutual inductance standard may be
represented as shown in Figure 6.18.
L
1
R
1
C
12
C
1
C
2
C
L
P
L
R
2
L
2
C
H
P
H
M
Figure 6.18 The equivalent circuit of a mutual inductor
The equivalent mutual inductance (M
E
) is then given by
M
E
M C
12
R
1
R
2
Mw
2
C
1
L
1
C
2
L
2
C
12
L
1
M L
2
M
M
_ _
6:23
where M is given by (6.20) and the signs refer to the seriesaiding or series
opposing connections, respectively. A typical relative correction to M at 10 Hz can
be of the order of 0.5 ppm. Similarly, the resistance of the mutual inductance
standard also has a frequency dependence and is given by
R
E
w
2
MC
1
R
1
MC
2
R
2
C
12
R
1
L
2
M R
2
L
1
M f g 6:24
which may be regarded as the leakage resistance of the mutual inductance and is
distinct from the much larger resistances R
1
and R
2
of the primary and secondary
windings, respectively. That is, for a given current in the primary winding, the
induced voltage in the secondary winding will depart, slightly, from the ideal 90
(i.e. voltage in quadrature with the current). This slight departure from ideal
quadrature voltage is equivalent to a voltage component that is inphase with the
current. Therefore, the ratio of the inphase voltage to the current is equivalent to a
resistance R
E
in (6.24), which is real and hence dissipative. This can be as small as
1 mW for a typical 10 mH mutual inductance standard at a frequency of 10 Hz.
An actual mutual inductor can easily be made so that it complies closely, at
low frequencies of the order of 50 Hz, with the conditions for an ideal mutual
inductor, that is, additivity and negligible departure from 90
1
k1
1
k
k
2
f
10
kbf
10
ka 6:27
C
L
P
L
C
H
P
H
resistance wire
shielding disk
and wire support
Figure 6.19 A Haddad coaxial resistance standard
Impedance standards 141
where e
0
8.85 10
12
F/m is freespace permittivity, a and b are the radii of the
resistance wire and the inner of the coaxial return conductor, R is the nominal value
of the standard at DC, and
f
10
kt
I
1
kpt=c=I
0
kpb=c K
1
kpt=c=K
0
kpa=c
I
0
kpa=c=I
0
kpb=c K
0
kpa=c=K
0
kpb=c
6:28
where I
0
(x), K
0
(x), I
1
(x), K
1
(x) are the modified Bessel functions of order zero and
one, respectively, and c is the length of the resistance wire. In general, expression
(6.27) can be equivalently rewritten as DR/R ~ w
2
C
2
R
2
, where C is the total dis
tributed capacitance of the coaxial standard. Endplate calculations were not given
but a rough estimate was made and for this standard at frequencies between 10 and
10 kHz the contribution of the end plates was not considered to be a major factor
(whereas in the HF calculable resistance standard to be described in section 6.5.3,
it is an important factor that should be accounted for). Haddad estimated the
change in resistance due to the distributed capacitance of the standards at 1.592 kHz
(or 10
4
rad/s) to be 6.9 10
14
and 5.2 10
12
for the 100 W and 1 kW resistance
standards, respectively. Furthermore, these values increase quadratically with fre
quency, as Haddads calculations at 15.92 kHz show. Similarly, the change in
resistance due to inductance of the resistance wire and the return coaxial conductor
is given by
L
0
m
0
2p
ln
b
1
a
_ _
2b
1
=b
2
1 b
1
=b
2
ln
b
b
1
_ _
3
4
lnz
_ _
6:29
where b
1
is the outer radius of the coaxial return conductor and the term ln(z) is a
function of the ratio (b
1
/b) tabulated in Grovers text book on inductance calcula
tions [15, Table 4]. The inductance given by (6.29) varies with frequency, such that
L
0
at DC approaches the highfrequency limit L
?
, given by
L
1
m
0
2p
ln
b
1
a
_ _ _ _
6:30
This inductance L (the limit of whose value is given either by (6.29) or (6.30))
leads to a change in the resistance R of the standards in two ways; first L combines
with R, such that DR/R ~ w
2
L
2
/R
2
, and second it combines with R and the dis
tributed capacitance C of the standards mentioned above, such that DR/R ~ w
2
C
2
L
2
.
The first factor leads to a resistance change of 2.1 10
11
and 2.2 10
11
for the
100 W and 1 kW standards, respectively, at 1.592 kHz. The second factor leads to a
resistance change of 1.1 10
9
and 1.6 10
11
for these standards at the same
frequency.
Haddad also considered the change in resistance due to eddy current losses in
the Evanohm (an Ni:Cr:Al alloy) resistance wire, copper support wires, copper
return conductor and the end plate. Only in the 100 W standard at 1.592 kHz did the
eddy current losses in the copper support wires and the copper return conductor
142 Coaxial electrical circuits for interferencefree measurements
result in significant resistance changes, of 8.9 10
10
and 5 10
10
, respectively.
For the copper support wires, the resistance changes quadratically with frequency,
whereas for the copper return conductor, it changes as w
3/2
. Furthermore, mineral
oil also makes a relatively small contribution to the change in resistance of the
standards because of two factors; first it increases the distributed capacitance
by a factor of its relative permittivity, e
r
2.16, and second its loss angle, tan d
3.5 mrad produces a distributed resistance between the Evanohm wire and the
copper return conductor, thus producing a relative change in resistance of DR/R ~
wRC tan d 2.2 10
12
and 1.9 10
11
for the 100 W and 1 kW standards at
1.592 kHz, respectively. Haddad also considered other factors that could influence
the resistance of the standards at 1.592 kHz relative to their DC values, such as
misalignment of the resistance wire from the centre of the coaxial return conductor
and unequal current distribution between the two split cylinders. These were esti
mated to contribute about 1 10
9
each to the resistance change. However,
Haddad regarded the contribution from mutual inductance between the voltage and
current leads to be negligible. In contrast, at high frequencies, the mutual induc
tance is expected to be significant, since its contribution depends quadratically on
frequency. Similarly, the endplate effects are also expected not to be negligible at
high frequencies. Both of these aspects are discussed in more detail in section 6.5.3
for the HF calculable coaxial resistance standard. Furthermore, Haddad also esti
mated the phase angle of the resistance standards as
j w CR
L
R
_ _
wt 6:31
where L (L
0
L
?
)/2 and t is the equivalent time constant of the standards. For
the 100 W and 1 kW standards, the calculated phase angles were estimated to be
33.6 0.8 and 9.4 0.6 mrad, respectively, at 1.592 kHz. As (6.31) indicates, both
vary linearly with frequency.
Finally, Haddad also compared these calculations with measurements of the
100 W and 1 kW standards, at 1.592 and 15.92 kHz, on a fourterminalpair bridge.
For the inphase change in resistance of the ratio of the 100 W and 1 kW standards,
he found a measured difference of (28.7 10) 10
8
between the two fre
quencies. The calculated difference was (20.3 3.7) 10
8
. Although these two
results agree within their stated uncertainties, Haddad noted that the relatively large
uncertainty in the measurements was due to two current equalisers not being as
effective as they should have been in their measurement setup. In contrast, Haddad
found a very good agreement between the calculated, 217.8 mrad, and measured,
225.6 mrad, phase angle change between the ratio of the 100 W and 1 kW standards
at the two frequencies.
Recently, Delahaye [16] at the International Bureau of Weights and
Measures (BIPM) used a 1290.6 W (equal to R
K
/20, where R
K
is the von Klitzing
constant) Haddad resistance standard to measure the frequency dependence of two
51.6kW resistance standards at frequencies between 500 Hz and 6 kHz. He then
measured with a quadrature bridge the frequency dependence of two sets of
AndeenHagerling 10 and 100 pF fusedsilica capacitance standards in terms of the
Impedance standards 143
frequency dependence of the 51.6kW resistance standards. The relative change in
the fourterminalpair capacitance value from 500 Hz to 6 kHz was a decrease of
between 70 and 120 parts in 10
9
for all four capacitance standards, with a standard
uncertainty ranging from 13 parts in 10
9
at 500 Hz to 34 parts in 10
9
at 6 kHz. The
measured data was fitted to a simple dielectric relaxation model that confirmed the
origin of the frequency dependence in the capacitance standards. The internal
inductances of the standards, which would normally lead to an increase in the
apparent capacitance of a standard at higher frequencies, were found to be no more
than 0.05 mH as confirmed by resonance frequency measurements. The corre
sponding frequency dependence of the dissipation factors was not reported.
Despite the elegance, simplicity and usefulness of the Haddad coaxial resis
tance standard for measurements at low frequencies, as it will be shown in section
6.5.3, this type of standard has several limitations at higher frequencies.
6.5.3 A nearly ideal HF calculable coaxial resistance standard
This standard was devised [17] because of the limitations of the Haddad (section 6.5.2)
[14] and Gibbings (section 6.5.5) [18] resistance standards at frequencies above
16 kHz. These limitations include the difficulty of accurately calculating the frequency
dependance of these standards above 16 kHz and the relatively large frequency
dependance (proportional to w
2
) resulting from the mutual inductance between the
current and potential leads. By identifying suitable solutions and improvements, a
nearly ideal calculable standard of resistance has been realised for the DCto 100 MHz
frequency range (see Figure 6.20). In contrast to the Haddad and Gibbings standard,
the measurement plane and the calculation plane are at exactly the same spatial
SMB connectors
crosssection
at SMB connectors
3
5
m
m
2
2
m
m
10 mm
resistance wire
HC
HC
LC
side elevation
removable (copper) end cap
outer copper shield
LP
LP
HP
HP
PTFE insulating layer
copper
return
conductor
Figure 6.20 The highfrequency 4TP calculable coaxial resistance standard for
broadband DC to 100 MHz operation ( IEEE 2005 [17])
144 Coaxial electrical circuits for interferencefree measurements
location and the inner fourterminalpair connections to the resistance wire and the
coaxial copper return conductor are orthogonal in order to minimise the mutual
inductance between the current and potential leads. The outer connections are to the
copper shield, and their orthogonality ensures that no outerconductor impedance is
added to the component. This configuration significantly reduces the f
2
frequency
dependence of the standard by minimising the mutual inductance between the leads
that is one of the main components of uncertainty in the MHz range of frequencies.
Supporting dielectric material is eliminated and there is only air between the resistance
element and the copper return conductor; any dielectric other than air would make a
finite additional contribution to its frequency dependence and complicate analytical
calculations. If the greatest achievable accuracy is required, the effect on the frequency
dependence of any oxide layers on the resistance wire element or on the inner surface
of the coaxial return conductor also must be calculated. In addition, there may be a
finite frequency dependence associated with the resistivity of the resistance element
(although this is expected to be quite small for f (f
p
, where f
p
is the plasma frequency
of the metal). End effects can also be virtually eliminated if the potential connections
are displaced by a distance comparable with the diameter of the inner return conductor
away from the current connections while maintaining the orthogonality of the four
terminalpair connections (as illustrated in the side elevation in Figure 6.20).
The performance of an HF calculable resistance standard can be represented by
a relatively simple model which contains only the main sources of frequency
dependence, although the typical accuracy of the calculated frequency dependence
at 100 MHz of either 100 W or 1 kW standards is only about 0.1% (1s). For most
applications, such as calibration of commercial fourterminalpair LCR meters and
impedance analysers, this is perfectly adequate, but for some metrological appli
cations, such as Johnson noise thermometry (where the 100W sense resistor needs
to be calibrated with about 1 ppm uncertainty at frequencies up to 1 MHz) or
quantum Hall resistance measurements to a few MHz, much more accurate ana
lytical calculations and measurements of a HF calculable resistance standard will
be needed in the future. This work is currently in progress and will be published in
due course. The major complications in calculating the frequency dependence more
accurately are the end effects at the end disc of the standard. Although this disc has
a calculated DC resistance of only 0.1 mW and the eddy current losses are unlikely
to change its resistance by more than a factor of 1000 from DC to 100 MHz (which
would amount to a contribution of only 0.1 ppm, for a 100 W standard), the effect of
the capacitive current distribution near the end disc and its inductance need to be
accounted for more accurately.
The impedance Z(w) of an HF calculable resistance standard can be described
in terms of a pure series resistance R
S
with a series inductance L
S
and an overall
shunt capacitance C
S
. This model is found to be suitably appropriate given the
geometrical design shown in Figure 6.20. The voltage between the return conductor
and shield is zero, and consequently, the capacitive current through the supporting
PTFE dielectric material is also zero. The impedance Z(w) of the standard is
Zw
R
S
jwL
S
1 w
2
L
S
C
S
C
S
R
2
S
1 w
2
C
2
S
R
2
S
2L
S
C
S
w
2
L
2
S
C
2
S
6:32
Impedance standards 145
Typical values for a 100W resistance standard, with the radius of the copper
return conductor r
1
10 mm and wire radius of r
w
15 mm and lumped para
meters L
S
and C
S
are approximately 0.2 mH and 1.8 pF, respectively. Similarly, for
a 1kW resistance standard, these are typically 0.3 mH and 1.3 pF, respectively.
Substituting these values into (6.32) shows that the 100 W standard has an apparent
positive coefficient of frequency dependence, that is, a 30ppm higher resistance at
1 MHz compared to its DC value. In contrast, the 1kW resistance standard has a
negative frequencydependent coefficient, that is, 40 ppm at 1 MHz compared to
its DC value. Furthermore, both resistance standards show a quadratic dependence
on frequency, with about 10% uncertainty of the frequencydependent coefficients.
The time constant t is also an important physical quantity of the resistance
standards and should be experimentally and theoretically accounted for in any
precision impedance measurements. From (6.32)
t
L
S
R
S
C
S
R
S
6:33
For the 100Wand 1kWresistance standards, for example, t &2 ns and t &1 ns,
respectively. Finally, the parallel equivalent circuit resistance of the standard is
related to the series equivalent circuit resistance by R
P
R
S
1 w
2
t
2
.
6.5.4 A bifilar resistance standard
This standard, which is suitable for frequencies less than 100 kHz, consists of a
resistance wire forming a single loop, of outward and returning current, surrounded
by a cylindrical shield. Figure 6.21 shows an equivalent electrical model of the
standard consisting of a uniform transmission line with shunt capacitance and
leakage conductance to the shield and between the two wires of the loop. For the
greatest accuracy, the standard should be terminated with fourterminalpair con
nections. Its conductance is
G
4TP
1
R
1
R
6
G
0
2G
1
w
2
L 2M
2
R
2
w
2
R
2
720
15C
2
0
C
0
4C
1
_ _
6:34
where R is the resistance of the standard, G
0
and G
1
are the conductances between
the resistance wire and the shield and between the wires of the loop, respectively, M
is the total mutual inductance between the two halves of the wire loop and C
0
and
C
1
are the shunt capacitances between the wires and the shield as well as between
the sides of the loop. The relative change in conductance, DG
4TP
1 G
DC
=G
4TP
where G
DC
is the conductance at DC, at a frequency of 1 kHz is approximately
1 10
9
for a 1kW resistance standard.
146 Coaxial electrical circuits for interferencefree measurements
P
H
C
L
P
L
C
H
G
1
C
1
C
0
C
0
G
0
G
0
Figure 6.21 Equivalent circuit lumped parameter model of a bifilar resistance
standard
6.5.5 Gibbings quadrifilar resistance standard
Initially, this standard [18] was devised for determining the ohm from the
ThompsonLampard calculable capacitor. This required a 10kW resistance
standard whose change in resistance between DC and 10
4
rad/s could be calcu
lated to within 1 part in 10
7
uncertainty. Many national standard laboratories
now employ this standard to relate the DC value of a resistor to its AC value at
w 10
4
rad/s. The DC value of a resistor is measured in terms of the quantum
Hall resistance (see also section 6.6.1) and comparison with the quadrifilar
standard at w 10
4
rad/s enables the impedance of the standard to be directly
compared to the impedance of a capacitance standard in a quadrature bridge (see
sections 9.2.19.2.3). For a 10kW value of the standard the change of resistance
with frequency is dominated by secondorder terms of the type w
2
R
2
C
2
. In con
trast, for resistance standards with lower values than 10 kW, the dominating
effects are of the type w
2
L
2
/R
2
, due to mutual inductive coupling and to dielectric
loss. For the intermediate nominal value of 1 kW eddy current losses that would
be induced in the shield conductor are cancelled by the reversed quadrifilar con
figuration of the resistance wire shown in Figure 6.22.
An additional advantage of the reversed quadrifilar configuration is that it
reduces the physical length of the standard, which is also advantageous for con
trolling its temperature. Gibbings in his original publication [18] reported both
100 W and 1 kW reversed quadrifilars whose change in resistance with frequency
from DC to 10
4
rad/s was less than 1 or 2 parts in 10
8
with the same uncertainty. A
10:1 ratio bridge scaled these values to the required 10 kW.
Impedance standards 147
The fourterminalpair impedance of the reversed quadrifilar resistance stan
dard can be defined as
Z
4TP
w R
0
1 a f jwt 6:35
where R
0
is the value of the standard at DC, and the dissipative or inphase com
ponent of the impedance a f is the summation of three contributions; b due to
reactive impedances, c due to eddy current losses induced in the shield conductor
and g due to skin effect in the resistance wires, thus
a f b c g 6:36
such that
b
R
0
6
G
0
5G
1
3G
2
w
2
R
2
0
11520
240C
2
0
15 C8C
1
8C
2
2
C
0
16C
1
2
_ _
w
2
R
2
0
L8M
1
4M
2
2
_
_
_
_
1
6:37
2a
Current and potential leads
Nylon screw with
ceramic washer
PTEE supporting rod
PTEE supporting rod
Resistance wire
s
h
V
H
V
3
V
2
V
1 V
4
i
4
i
1
i
2
i
3
I
H I
L
V
L
2d
2b
2c
w
2
R
2
0
737280
192C
2
0
4 C
0
32C
1
32C
3
2
60 C
0
16C
1
32C
2
16C
3
2
15 C
0
16C
1
16C
2
16C
3
16C
4
2
_
_
_
_
w
2
R
2
0
L 16M
1
16M
2
16M
3
8M
4
2
_
_
_
_
6:42
where the symbols and subscripts have the same meanings as in (6.37). Its time
constant is
t
R
0
192
17C
0
128C
1
144C
2
128C
3
48C
4
1
R
0
L 16M
1
16M
2
16M
3
8M
4
_
_
_
_
6:43
Similarly, the eddy current losses induced in the shield conductor are given by
c
32m
0
p
m f wh
R
0
1
16 m
2
f
_ _
b
a
_ _
8
6:44
where m( f ) is given by (6.39), and the remaining factor in (6.36) for the eddy
current losses in the resistance wire is
g
w
2
192
m
0
m
r
r
2
r
w
_ _
2
6:45
where r
w
and r
w
are the radius and resistivity of the resistance wire, respectively.
The quadrifilar and octofilar resistance standards assisted in initially verifying
the frequency independence of quantum Hall resistance devices (quantised at
12.906 or 6.45 kW) at frequencies ranging up to 5 kHz. Typically a 12.906kW
quadrifilar standard changes fractionally with frequency by approximately 1 10
7
to 2 10
7
at 5 kHz and has a time constant of 1015 ns. An octofilar standard
changes fractionally with frequency by 0.55 10
7
at 5 kHz and has a time constant
of 10 ns. Both have an uncertainty of approximately 3 10
8
(1s) at 5 kHz.
6.5.7 HF secondary resistance standards
The preceding sections discuss standards of resistance whose frequency depen
dence is calculable from first principles. There is also a need for secondary stan
dards whose frequency dependence, although not calculable, is nevertheless as
small as possible. A simple and practical fourterminalpair secondary impedance
standard with applications in highfrequency metrology is shown in Figure 6.23.
Four female BNC connectors are mounted equidistant along the lid of a 70 mm
20 mm 20 mm diecast box so that their spacing allows a direct plugin to the
input ports of commonly available commercial impedance bridges and network
analysers. The outer conductor of the L
C
connector makes contact with the metal of
150 Coaxial electrical circuits for interferencefree measurements
the box, and the other three connectors are insulated from it so that the box con
stitutes an electrical shield. The inline configuration of the BNC connectors is
converted to an orthogonal geometry of current and potential connections as
shown, with the inner conductors connected to a surfacemount chip that can be a
resistor, a capacitor or an inductor. These chips are widely available from numerous
manufacturers of electronic components. The connections are made with coaxial
cables whose outer conductors are brought together as a star connection adjacent to
the chip. This arrangement minimises the mutual inductance between potential and
current connections. For resistance or inductance standards, the chip penetrates a
closely fitting hole in a planar shield connected to the star point to minimise direct
capacitance from one end to the other. This reduces the phase angle and frequency
dependence of the standard. Its frequency response can be calibrated using an HF
calculable resistance, capacitance or inductance standard. This design of resistance
standard is a good highfrequency transfer standard for calibrating highfrequency
fourterminalpair meters and impedance analysers. If the standard is connected to
an instrument via short coaxial cables rather than being directly plugged in, a cable
impedance correction to the measured results is normally required, and in some
cases, cable loading corrections may also be needed if the measurement accuracy
for a particular application demands it. These effects are typically of the order of
0.1% or less at 1 MHz.
An alternative construction based on four female SMB connectors soldered
into a copper plate on the corners of a square to ensure an orthogonal geometry is
illustrated in Figure 6.24. The resistance, capacitance or inductance chip component
surfacemount resistor
with star point below
shield
L
C
H
C
L
P
H
P
directly to LCR meter
locklever and plugonly
BNC connectors
20 mm
coaxial leads
inner shield
star point
SMD
(a)
(b)
(c)
Figure 6.23 (a) Highfrequency secondary 4TP resistance standard using a
(b) internal coaxial transmission lines (c) and surfacemount
chip resistor. The details (b) and (c) show the coaxial leads before
and after mounting the chip
Impedance standards 151
is joined to the inner conductors of the SMB connectors in the orthogonal geometry
as shown. For resistance or inductance standards constructed in this way, the chip
component is also mounted through a closely fitting hole in a shield between the high
and low SMB terminals. This minimises its shunt capacitance and, consequently,
reduces its phase angle and frequency dependence. Surfacemount chip resistors are
available in values ranging from 1 W to 1 MW, and can be obtained with values
within 0.01% of nominal. The temperature coefficient of the surfacemount chip
component is typically not more than 10 ppm/
z
2 2
p
pas
Bei
1
az Ber
0
azBei
0
az Ber
1
az Ber
0
azBei
0
az
Bei
1
az
2
Ber
1
az
2
_ _
6:47
where R
i
S
is the series resistance of the inner resistance wire per unit length,
z H2/d H(wms), d is the skin depth, s is the conductivity of the resistance wire
and Ber
n
and Bei
n
are the complex Kelvin functions of order n. The expression in
(6.47) is plotted in Figure 6.30, which shows the calculated frequency dependence
of a practical 1kW resistance wire in an HF calculable resistance standard. The key
physical properties of this standard needed for calculating its electromagnetic
properties are given in Table 6.2. The calculated frequency dependence due to eddy
currents in the resistance wire at frequencies up to 100 MHz is plotted on the
bottom of Figure 6.30 and shows that the apparent resistance of the wire increases
quadratically with frequency. The total change in resistance, from DC to 100 MHz,
is found to be higher by only 8.2 ppm. in contrast, the linearlog plot shown on the
top of Figure 6.30 shows that the skin effect becomes significant as the frequency
1.2 10
3
2.0 10
7
0.0 4.0 10
7
frequency/Hz
frequency/Hz
6.0 10
7
8.0 10
7
1.0 10
8
977.595
977.592
977.589
977.586
10
2
10
4
10
6
10
8
10
10
R
/
R
/
1.6 10
3
2.0 10
3
2.4 10
3
Figure 6.30 Calculated frequency dependence from equation 6.47 of the ideal
1kW HF calculable resistance standard up to 100 MHz (bottom) and
up to 100 GHz (top)
Impedance standards 159
approaches 5 GHz, and the frequency dependence has a Hf dependence, as
expected for eddy current losses in the skineffect regime. Furthermore, neglecting
radiation loss effects, calculations using the software Maple have also confirmed
that as f ? 0, (6.44) gives the DC value of the standard, as shown in Figure 6.30.
The selfinductance of the 1kW resistance wire is
L
i
S
z
2 2
p
pasw
Bei
1
azBer
0
az Bei
0
az Ber
1
azBer
0
az Bei
0
az
Bei
1
az
2
Ber
1
az
2
_ _
6:48
where L
i
S
is the series selfinductance of the resistance wire per unit length,
assuming that the current is returned and the circuit completed via a coaxial return
conductor at infinity. Figure 6.31 shows the calculated frequency dependence of the
selfinductance. The bottom graph shows that the inductance for the actual finite
sized standard is essentially constant at approximately 8 nH for frequencies up to
Table 6.2 The physical parameters of a 1kW HF calculable resistance standard
for calculating the results shown in Figure 6.30
Nominal value r a b c Length
1 kW 1.33 mW.m 0.00835 mm 10 mm 11 mm 161 mm
The resistivity of the resistance wire is and its radius is a; b and c refer respectively to the inner and
outer radii of the copper return conductor.
frequency/Hz
10
8
10
9
10
10
10
11
3.0 10
7
6.0 10
7
9.0 10
7
0.0
frequency/Hz
8.0498
i
n
d
u
c
t
a
n
c
e
/
n
H
i
n
d
u
c
t
a
n
c
e
/
n
H
8.0499
8.0500
8.0501
8.0502
3
4
5
6
7
8
Figure 6.31 Calculated frequency dependence of the selfinductance of the
1kW HF calculable resistance standard at frequencies up to
100 MHz (bottom) and 100 GHz (top). The erratic lowfrequency
behaviour is merely an artefact of the numerical calculations
160 Coaxial electrical circuits for interferencefree measurements
100 MHz because the skin effect in the 16mm diameter resistance wire is negli
gible up to 100 MHz. In contrast, for frequencies above 5 GHz, when the skin effect
in the wire begins to become prominent, as shown in the top graph, the apparent
selfinductance reduces with a dependence proportional to 1/Hf, as expected.
The section of coaxial return conductor that contributes to the actual defined,
calculated and measured impedance is only that between the two planes located at
H
P
and L
P
. That is, between these two planes is an active section of the coaxial
return conductor that contributes to the defined impedance in terms of losses,
inductance and shunt capacitance. This is because since in this ideal geometry end
effects are negligible, the circumferential magnetic flux lines caused by the trans
ported AC in the inner resistance wire that cut the outer coaxial return conductor
only contribute to the measured impedance between the H
P
and L
P
connections, as
required by the fourterminalpair definition. Therefore, it is possible to calculate,
again from first principles using Maxwells equations, the frequency dependence of
the eddy current losses and selfinductance of this section of the return conductor.
In the case of eddy current losses,
R
0
S
z
2
p
I
pk
faBer
0
cz Bei
0
cz bKei
0
cz Ker
0
czg 6:49
where R
0
S
is the series resistance of the outer coaxial return conductor per unit
length. I
pk
is the peak AC transport current in the inner and outer conductors and a
and b are parameters, also involving the complex Kelvin functions defined in
Appendix 4.
The corresponding frequency dependence of the selfinductance of the active
part of the outer return conductor is
L
0
S
z
2
p
wI
pk
aBer
0
cz Bei
0
cz bKer
0
cz Kei
0
cz 6:50
where L
0
S
is the series selfinductance of the outer return conductor per unit length.
Although similar expressions to those of (6.47) to (6.50) can be found in the
literature, they are often full of errors and do not give consistent results. Therefore,
to the best of our knowledge, these are the first set of correct expressions that have
also been robustly checked for consistency using Maple. By robust consistency we
mean first that as the frequency f ?0, the above expressions lead to the DC value
of the resistance standard, as shown in Figure 6.30, and second, at frequencies
where the skin effect is negligible, the frequency dependence of the resistance wire
should be proportional to f
2
due to eddy current losses in the wire, as demonstrated
in Figure 6.30 (lower graph). Finally, when the skin effect in the wire is present
then the frequency dependence of the resistance should be proportional only to Hf,
which is also shown in Figure 6.30 (upper graph). Similar robustness checks have
also been carried out on the calculated selfinductance of the inner resistance wire
and the outer coaxial return conductor.
Impedance standards 161
The influence of shunt capacitances on the apparent resistance of the standard
is as expressed in equation 6.27 of section 6.5.2 when normalised by w
2
R
2
.
6.6 Quantum Hall resistance
6.6.1 Properties of the quantum Hall effect (QHE) and its use as a
DC resistance standard
The integer QHE discovered in 1980 by Klaus von Klitzing [22,23] is a macro
scopic quantum effect and its extraordinary and fascinating precision makes it very
valuable and important for resistance and impedance metrology. The QHE occurs
in a socalled twodimensional electron gas where the electrons of a thin planar
layer in a semiconductor are confined to twodimensional motion. This can be
achieved with a GaAs heterostructure semiconductor device that has a doped layer
designed in such a way that the electrons of the donor atoms accumulate in a very
thin layer and so constitute the twodimensional electron gas (Figure 6.32). A
typical quantum Hall device manufactured from such a heterostructure is shown in
Figures 6.33 and 6.34.
Usually, the twodimensional electron gas is formed as a rectangular bar with a
current terminal at each short side and three potential contacts along each long side.
The quantised state occurs only in a strong perpendicular magnetic field that forces
the electrons to move in cyclotron orbits with quantised energy levels (the socalled
Landau levels). To avoid a thermal breakdown of the quantised state, the tem
perature of the QHE device must be below a few Kelvin, and the current must be
below a certain threshold (as discussed below). Under this condition and measured
with DC, the Hall resistance defined as
R
H
U
H
I
GaAs cap layer
undoped AlGaAs
undoped GaAs
semiisolating GaAs
2DEG
13 m
1040 nm
1020 nm
3050 nm
ca. 400 m
E
Z
Sidoped AlGaAs
E
F
Figure 6.32 The layered structure of a quantum Hall device
162 Coaxial electrical circuits for interferencefree measurements
shows ranges of the applied magnetic induction B over which R
H
is constant. That
is, quantised steps, which are also termed plateaux, occur successively in a graph of
R
H
versus B according to
R
H
R
K
i
B
2DEG
contact pads
wafer
I
I
U
xx
U
H
Figure 6.33 The terminals and layout of a quantum Hall bar
Figure 6.34 A photograph of a quantum Hall device mounted on a printedcircuit
board, which is 16 mm in diameter and designed in the frame of the
Euromet project No. 540
Impedance standards 163
with R
K
h/e
2
, the von Klitzing constant, and i 1, 2, 3, . . . an integer quantum
number. The electron density of the twodimensional electron gas determines the
magnetic induction (typically of the order of 10 T for the most commonly measured
i 2 plateau), and the mobility and the bath temperature determine the width of the
plateaux.
The longitudinal resistance, defined as
R
xx
U
xx
I
always vanishes for a fully quantised state of the twodimensional electron gas. The
Hall resistance of a poor device can deviate from the quantised value, but this is
always accompanied by a nonzero longitudinal resistance, and so the measurement
of the longitudinal resistance is a key parameter for testing whether the quantisation
of a particular device is complete. A typical measurement of the quantum Hall
and longitudinal resistances as a function of magnetic induction is shown in
Figure 6.35. From the various welldefined plateaux available for measurement, the
quantum Hall resistance at the plateau i 2 whose value of 12.906 kW is closest to
a decade value, occurs in devices with a practicable electron mobility at a magnetic
induction that is easily produced by a superconducting magnet. This plateau is
particularly wide and is therefore commonly used in resistance metrology. The
detailed requirements for reliable DC quantum Hall resistance measurements as
well as for suitable devices are specified in the DC guidelines document [24,25]
and are not repeated here.
0
0 2 4 6
B/T
i = 4
i = 3
i = 2
8 10
2000
4000
6000
R
H
/
R
x
x
/
8000
10000
12000
0
300
600
900
1200
1500
1800
T = 0.3 K, I = 10A
Figure 6.35 The quantum Hall and longitudinal resistance of a GaAs
heterostructure
As far as it is known, the von Klitzing constant R
K
depends only on two
unchangeable fundamental constants, Plancks constant h and the charge of the
electron e, in the combination h/e
2
. It does not depend on any other parameter of
our macroscopic world, such as the geometrical dimensions of the twodimensional
electron gas, the semiconductor material, time, temperature, barometric pressure or
164 Coaxial electrical circuits for interferencefree measurements
humidity. It does not require any precise geometrical adjustment or dimensional
measurements, it does not depend on all the side effects that occur with conven
tional resistance artefacts, and it is neither subject to longterm drift nor affected by
transportation, unlike conventional resistance standards. Therefore, the quantised
resistance value can be reproduced with an impressive relative uncertainty of at
least 10
10
, which is much better than all conventional resistance artefacts, and this
extraordinary property is what makes quantum Hall resistance so valuable for
resistance and impedance metrology.
The physics that governs this quantum effect is outlined in several other pub
lications. Although some details are still not completely understood, for metrolo
gical purposes, the important facts are well supported by experiment and are
sufficient to establish the utility of a quantum Hall device as an excellent DC
resistance standard [2325].
The only possibility for accurately measuring the SI value of the von Klitzing
constant without assuming any relation to other fundamental constants is the
measuring chain that links a calculable capacitor to the quantum Hall resistance
(Figure 6.36). But its SI value can also be obtained from the measured value of
the fine structure constant. SI resistance and capacitance can then be obtained from
the inverted measurement chain of Figure 6.37. The individual measuring chains
that have been realised up to now in various NMIs follow a scheme similar to either
Figure 6.36 or Figure 6.37. Their implementations differ in some details, but they
all require quadrature and 10:1 ratio bridges, a calculable ACDC resistor to effect
the ACDC transfer (see sections 6.5.16.5.4), and either a DC resistance ratio
bridge or a cryogenic current comparator. All of these components together with
calculable capacitors are described in this book.
Now the alternative of a direct AC measurement of the quantum Hall resis
tance is not only possible but offers an even better uncertainty via the shorter
measurement chain of Figure 6.38 [26].
Following a recommendation of the Comite International des Poids et Mesures
that has been in force since 1 January 1990, a conventional value, R
K90
25 812.807 W, has been chosen as the best representation of the SI value of the von
Klitzing constant. The choice was made from a consensus mean value of the results
that had been obtained from calculable capacitors. R
K90
is a defined, exact quan
tity and the resulting DC resistance unit, the W
90
, has no uncertainty according to
this agreement. It is not exactly the same as the SI unit and in SI units the W
90
has a
relative uncertainty of 2 10
7
.
A second numerical value of the von Klitzing constant recommended by the
CODATA committee [2] assumes the relation R
K
h/e
2
to be exact, an assumption
that is justified by theory as well as by experimental universality tests. These have
been verified with a very high precision of about 10
11
that R
K
does not depend on
any other quantity. The numerical SI value of R
K
with a relative uncertainty of
6.8 10
10
can thus be inferred from an experimental value of the fine structure
constant, a (m
0
c
2
/2)e
2
/h. The constants m
0
(permittivity of vacuum) and c (speed
of light in vacuum) contribute no extra uncertainty as their values are defined to be
exact in SI units. The SI units of resistance, capacitance and inductance can also be
Impedance standards 165
realised from this value of R
K
. Its very low uncertainty neither limits the accuracy
of any resistance or impedance calibration nor can it be made better by any present
day calculable capacitor. Nevertheless, resistance and impedance standards so far
are required to be calibrated in terms of R
K90
.
6.6.2 The properties and the equivalent circuit of a
quantum Hall device
As noted in the previous section, in the quantised state, the Hall resistance measured
with a Hall bar having current terminals at either end and potential terminals in the
12.9 k
12.9 k
1.29 k
quadrature bridge
frequency
10 nF
10 nF
DC QHR
CCC
ACDC transfer
1 pF calculable capacitor
10:1 bridge
10:1 bridge
10:1 bridge
10:1 bridge
1 nF
100 pF
10 pF
10:1 bridge
10:1 calibration
Figure 6.36 A measuring chain from a calculable capacitor to the DC quantum
Hall resistance
166 Coaxial electrical circuits for interferencefree measurements
middle of the device and at either side of it is an integer fraction R
H
R
K
/i of the von
Klitzing constant, with i 1, 2, 3, . . . , and the longitudinal resistance is zero. In
the following, we consider only ideal QHE devices, which means, for example, that
the dimensions of the twodimensional electron gas are sufficiently large and that the
contact resistances and the Hall current are sufficiently low, as specified in the DC
guidelines of the QHE [24,25]. Contact with the twodimensional electron gas can be
made by diffusing gold or tin into it. Gold contacts are usually laid down through a
lithographic mask and are geometrically quite precise. Tin in the form of sub
millimetre diameter balls can also be diffused in by heating, and good contacts made
this way can be of very low resistance but are inevitably geometrically less precise.
An ideal quantum Hall device has some basic and unique properties:
1. The twoterminal resistance measured between any pair of terminals is, apart
from the resistance of the measuring leads and the particular contact resis
tances involved, equal to the quantised Hall resistance value.
12.9 k
DC QHR
1.29 k
12.9 k
CCC
10:1 bridge
10:1 bridge
10:1 bridge
10:1 bridge
ACDC transfer
quadrature bridge
frequency
10 nF
10 nF
100 pF
10 pF
1 nF
10:1 calibration
Figure 6.37 A measuring chain from the DC quantum Hall resistance to a 10pF
capacitance standard
Impedance standards 167
2. When two different potentials are applied to any pair of terminals of the
quantum Hall device so that a current flows through the device, the two
device edges from the respective current entry and exit terminal in a clockwise
direction as seen against the direction of the magnetic induction are at the
same potential as the particular current terminal.
3. When a fourterminal measurement is carried out, the Hall voltage between
any pair of potential terminals on opposite sides of the current flow gives the
same quantised value.
4. The equipotential effect described under (2) of the list has a further con
sequence. In general, the metallic current terminal pads in contact with the
edge of the twodimensional electron gas have a finite width. A current entering
or exiting the twodimensional electron gas causes the equipotential effect also
along these contact borders but can enter or exit the twodimensional electron
gas only in a region having a potential difference between it and the contact.
This means that the currents entering or exiting regions automatically bunch
to the most extreme counterclockwise corner of the contact border as seen
against the direction of the magnetic induction. The full potential difference
occurs at each of these corners. Also the whole energy dissipation I
2
R
H
occurs there and can be made visible, for example, by imaging of cyclotron
emission [27] whereas the rest of the twodimensional electron gas is noise
and dissipationless. Therefore, the regions at the corners are often called
hot spots.
10:1 bridge
10:1 bridge
10:1 bridge
10:1 calibration
AC QHR
AC QHR
quadrature bridge
frequency
10 nF
10 nF
1 nF
100 pF
10 pF
Figure 6.38 A measuring chain from an AC measurement of quantum Hall
resistance to a 10pF capacitance standard
168 Coaxial electrical circuits for interferencefree measurements
After the hot electrons have entered the twodimensional electron gas
through the hot spot and assuming that the distance between the hot spot
and the defining potential terminals is large enough, the hot electrons ther
malise to the quantised state and are unaffected by all the complicated physics
of the hot spot. The electrons flowing through the twodimensional electron
gas need a finite time for this transition, and this makes clear that a hot spot has
a finite dimension (of the order of some 10100 mm). It is not a singularity
although it is often termed so in the literature. The same situation in reverse
applies to the hot spot at the opposite corner of the device where the electrons
accumulate before they leave the twodimensional electron gas by tunnelling
into the current contact. If the voltage across the device is too large, avalanche
effects of ballistic electrons hinder the development of the quantised state.
This breakdown typically happens at a few hundreds of microamperes and
appears as a sudden and dramatic increase of the longitudinal resistance by
several orders of magnitude [28].
If the current contacts are split into several parallel subcontacts with
appropriate series resistances (i.e. multiples of R
H
), the current can be shared
between them and the hot spot is divided into several hot spots (each with less
energy dissipation), but in practice, this approach gives little relevant
improvement of the device performance.
The fact that the effective point of contact lies in the extreme corner of
each current contact applies also to the potential contacts. As a result, the
direction in which the Hall voltage is measured differs a little from the geo
metrical direction of the potential contacts (Figure 6.33). Even if the potential
contacts are exactly opposite to each other, the longitudinal resistivity is
superposed on the quantum Hall resistance. For precision lithographic con
tacts, the proportionality parameter (the socalled sparameter) often is equal
to the width of the potential arms divided by the width of the twodimensional
electron gas and has a typical value of 0.1, but other and positive values have
been observed. The sparameter of tinball contacts usually has larger, nega
tive or positive values because of their unavoidable geometrical imperfection.
A negative sparameter produces a weak minimum at the highfield side of the
quantum Hall plateaux, and a positive sparameter produces a weak maximum
at the lowfield side of the quantum Hall plateaux. These effects occur because
as the magnetic induction is changed to move away from the plateau centre,
R
xx
increases. The Hall resistance further from the plateau centre also deviates
from the quantised value R
H
and at a greater rate than R
xx
. As a result, R
xx
causes a structure in a limited range of magnetic induction at the edges of the
Hall plateau. Further, if the longitudinal resistance is not zero across the whole
central plateau region, the shape of the quantum Hall plateaux changes
accordingly.
5. When a small current I
0
flows through a potential terminal (e.g. a capacitive
AC current, a thermoelectric DC current, or a deliberately injected current to
compensate for some unwanted voltage drop in the QHE device), it comes
from the preceding current terminal in counterclockwise direction as seen
Impedance standards 169
against the direction of the magnetic induction and causes a potential drop I
0
R
H
between the ends of the edge joining the current and potential terminals.
Because a QHE device behaves as a linear network, the effect of currents
simultaneously flowing through different terminals can be calculated accord
ing to the principle of superposition. AC QHE measurements often make use
of this property, as will be shown later on.
In an AC measurement, a conductor connected and bonded to a high
potential contact of a quantum Hall device should not be left unconnected at its
other end because the associated capacitive current would be drawn through
the quantum Hall resistance device and the Hall voltage of that current would
falsely contribute to the measured Hall voltage. Therefore, openended con
ductors are not permitted in AC quantum Hall resistance measurements.
Moreover, when a DC quantum Hall resistance measurement is carried out and
the current direction is reversed to eliminate thermal emfs, the capacitance of
an openended highpotential conductor is recharged, and due to dielectric
relaxation, the effect of charging may last several minutes before a reliable
Hall resistance can be measured.
To facilitate mathematical analysis and to support the interpretation of experi
mental results, a phenomenological equivalent circuit of a quantum Hall resist
ance device with four terminals has been suggested by Ricketts and Kemeny [29].
Their equivalent circuit is shown on the lefthand side of Figure 6.39. It consists of
linear circuit elements that are ideal resistors and currentdriven voltage sources
and is constructed symmetrically in such a way that neglecting the resistance of
the leads and contacts the twoterminal DC resistance between any two terminals
gives the quantised resistance value R
H
(property (1) from the list above).
The alternative equivalent circuit shown on the right of Figure 6.39 can be
derived from the RickettsKemeny scheme using network theory and is thus fully
equivalent to it [30]. It requires fewer components and is thus more suitable for
mathematical analysis as well as aiding intuitive understanding. To describe a QHE
I
2
I
2
I
3
I
3
I
4
I
4
I
1
I
1
R
H
/2
R
H
/2
R
H
/2
R
H
/2
B
B
(I
1
+ I
2
)R
H
/2
I
1
+ I
2
+ I
3
+ I
4
= 0
(I
2
+ I
4
)R
H
/2
I
2
R
H
I
3
R
H
R
H
(I
1
+ I
3
)R
H
/2
(I
3
+ I
4
)R
H
/2
Figure 6.39 The equivalent circuit of Ricketts and Kemeny and an equivalent,
simpler circuit
170 Coaxial electrical circuits for interferencefree measurements
device with two current terminals and two (or three) potential terminals at each side
of the device, this alternative scheme can be easily extended (Figure 6.41b) and
enables the basic properties of a QHE device listed above to be clearly visualised.
For the same situation, the RickettsKemeny scheme has to be joined up three
times in series but the many components and relations make the analysis quite
complicated. Furthermore, the alternative scheme, but not the RickettsKemeny
scheme, permits longitudinal resistances that can be different at either side of the
device to be added to the equivalent circuit.
In all equivalent circuits of QHE devices that are shown in this book, the
voltage sources are for convenience represented as alternating voltage sources but
of course, all considerations also apply to DC.
6.6.3 Device handling
QHE devices should be cooled down slowly from room temperature to 4 K or
below at a constant rate of not more than about 10 K/min. They should be cooled in
the dark, with no applied voltages or currents, with the connections to them short
circuited, and be shielded from radiofrequency radiation. Once cooled, the mea
suring leads should be handled carefully to prevent the device from being exposed
to static electrical charge. The measuring current should be turned on and off
smoothly as otherwise a large current pulse could flow through the device and
affect it, probably because of frozenin charge carriers. Frozenin charge carriers
(usually electrons) can affect the internal capacitances and their associated loss
factors resulting in irregular features appearing in AC plateaux and causing error in
AC measurements. The magnitudes of these features are proportional to frequency,
so they do not appear in DC measurements, but they nevertheless affect the desir
able spatial orthogonality of current path through the device and the potential
across it (the DC sparameter), and therefore may affect the quantised DC value. It
is also good practice to shortcircuit the leads of a quantum Hall resistance device
when no measurement is being carried out and when the magnetic field is rapidly
ramped up or down. Devices intended for AC measurement should be exposed to
a sufficiently good vacuum with mild baking at a maximum of 60
C for about
48 hours to remove electrically lossy water films before cooling.
These effects of misshandling are cumulative, but fortunately nature is kind in
that a device can always be returned to its original good state by simply warming it
up to room temperature and cooling it again.
6.7 QHE measured with AC
The QHE measured with DC has been well established since 1990 for making
precise DC resistance calibrations. Because of this great success, Melcher et al.
suggested [31] that quantum Hall resistance could be measured with AC at fre
quencies in the kHz range and the same quantum effect could then be used for AC
impedance calibrations.
Impedance standards 171
A measurement of the quantum Hall resistance with alternating current is
practically the same as the QHE measured with direct current, and the commonly
used phrase AC QHE does not imply a new physical phenomena it is only a
short form of QHE measured with AC. One small effect of AC measurement
arises because the localised electrons that are essential for the finite width of the
plateaux oscillate cyclically in the alternating electric Hall field. This gives rise to
an intrinsic frequency dependence that might be relevant in the GHz frequency
range. In the kHz frequency range, it is smaller than 10
9
R
H
/kHz and may also be
negligible for frequencies up to 1 MHz because the achievable uncertainties of
actual measurements are larger than for audiofrequency measurements.
Furthermore, while in a DC measurement of quantum Hall resistance, the
capacitances associated with a quantum Hall device are charged only during a short
time interval after switching the Hall current on or off or after every reversal of the
current direction, they are continuously charged and recharged in the case of an AC
quantum Hall resistance measurement. These capacitances have associated current
and frequencydependent AC losses that vary across a quantum Hall plateau. As a
result, in the early measurements, the quantum Hall plateaux measured with AC
were not as flat as those measured with DC, the AC quantum Hall resistance dif
fered from the quantised DC value, and both effects increased linearly with fre
quency and current. These effects limited the accuracy of AC quantum Hall
resistance measurements before special precautions were developed to compensate
or remove the effects of AC loss currents flowing through the device.
To avoid these problems that are less than 1 ppm/kHz but are nevertheless
significant, a few NMIs have traced their impedance standards to DC quantum Hall
resistance measurements by employing a special resistance artefact that has a cal
culable difference between its value measured with AC and with DC (its ACDC
difference) to effect the transfer between AC and DC. A possible measuring chain
is shown in Figure 6.36 or Figure 6.37. Other measuring chains using standards of
other or nondecade nominal values are also possible. Since then all the early dif
ficulties of AC quantum Hall resistance measurements appear to be understood and
solved (as will be reported in sections 6.7.4 and 6.7.5), and it is possible to measure
the QHE directly and precisely with alternating current. The capacitance unit, the
farad, can now be realised
by a short measuring chain (Figure 6.38),
from an unchangeable and highly reproducible cryogenic quantum effect,
with good resolution because of a very low thermal noise power,
without the need for any artefact of calculable frequency dependence that
might suffer from other limitations,
and therefore with a remarkably low relative uncertainty that is less than
1 10
8
.
As inductance is usually derived from capacitance and AC resistance, the three
impedance units ohm, farad and henry, as well as the ohm defined for DC can be
derived from the same quantum effect. This is an advantage for the consistency of
the SI.
172 Coaxial electrical circuits for interferencefree measurements
6.7.1 Multipleseries connection scheme
The conventional fourterminalpair measurement scheme cannot be applied
straightforwardly to an AC quantum Hall resistance measurement. Unlike all con
ventional fourterminalpair standards, the potential terminals of a quantum Hall
device do not have a lowimpedance connection to the associated current terminal.
Consequently, capacitances and inductances associated with the quantum Hall
device and its connections and surroundings give rise to serious frequency
dependent systematic errors. A multipleseries connection scheme [32,33] avoids
all these difficulties in an elegant manner and thus became the standard connection
scheme for AC measurements of quantum Hall resistance. This scheme requires
that the inner conductors of coaxial leads from all equipotential contacts on either
side of the quantum Hall device are connected to two respective star points. An
example is shown in Figure 6.40. The coaxial star points are usually outside the
cryostat and at room temperature. The two star points are the internal defining
points of a twoterminalpair device but, as will be shown later, with the advantage
that the cable corrections are numerically the same as, but are of opposite sign to
the simpler ones of a fourterminalpair standard.
cryostat
B
Z
H
= U/I
U = 0 U
I
I
I 0
I 0
Figure 6.40 A tripleseries connection
The multipleseries connection scheme is not restricted to AC quantum Hall
resistance measurements. It can also be used for DC quantum Hall resistance
measurements. Because shielded leads are also advantageous for DC measurements
(e.g. they reduce the requirements for the insulation resistance of the leads) and
because the multipleseries connection scheme also eliminates unwanted DC
effects such as the slow relaxation of charge storage in the dielectric insulation of
the measuring leads, one and the same quantum Hall device in one and the same
cryomagnetic system can be used for both DC and AC measurements.
As an example, Figure 6.42 shows the equivalent circuit of a tripleseries
connection scheme. To explain the multipleseries connection scheme, we start
with a twoterminal measurement at the two current terminals. The measuring
current undergoes unwanted voltage drops at the lead and contact resistances to the
current contacts so that the Hall potential at each edge of the quantum Hall device
differs from the potential applied to the room temperature end of its corresponding
current lead. If now another lead is connected from each star point to the next
potential terminal around the device, as drawn in Figure 6.42 the small potential
Impedance standards 173
difference between each edge and the respective star point causes a small current to
be drawn through the potential lead. This current causes an additional Hall voltage
that compensates for the small voltage drop along the current lead. The compensation
is not complete because this small current causes an even smaller voltage drop down
the added lead but the process can be repeated by another connection to the next
potential terminal. When one additional lead after another is connected from each
star point to the successive potential terminals down either side of the device, the
currents in them become successively smaller. If the number of leads is large enough
and the lead and contact resistances are small enough, the Hall voltage between the
two final potential contacts on either side of the device becomes practically the same
as the voltage applied to the two accessible star points outside the cryostat, which is
the purpose of the multipleseries connection scheme. This compensating effect can
be analysed with the aid of the equivalent circuit of Figure 6.42 [30].
(a)
High
Low
R
xxLo
R
xy
R
xy
R
xxHi
B
x
y
d
w
w
P
hot spot
(b)
I
3Lo
I
2Lo
I
1Lo
I
2Hi
R
H
I
1Hi
I
2Hi
I
3Hi
I
3Hi
R
H
I
3Lo
R
H
I
2Lo
R
H R
H
2DEG
equipotential edge
B
Figure 6.41 (a) The relevant device edges of a sixterminal device
(b) The equivalent circuit (for simplicity without the longitudinal
resistances)
The relevant lead impedances and contact resistances are shown explicitly in
Figure 6.42. For simplicity, the longitudinal resistances down each side of the
device, which should be zero if it is fully quantised, have been neglected in the
following analysis. The multiple leads at the high and the low side are labelled n
and m, respectively. The apparent impedance Z
H
as defined at the high and low star
points is calculated with respect to the quantum Hall resistance R
H
and yields to a
good approximation:
174 Coaxial electrical circuits for interferencefree measurements
Z
H
R
H
1
Z
nHi
Y
nHi
2
n
k1
~
Z
kHi
R
H
_ _
_ _
1
Z
1Lo
Y
1Lo
2
m
k1
~
Z
kLo
R
H
_ _
_ _
6:51
with Z
k
R
k
jw L
k
6:52
Y
k
wC
k
j tan d
k
6:53
~
Z
k
Z
Ck
Z
k
1 Z
k
Y
k
% R
Ck
R
k
(R
H
6:54
The first indices (k, l, m or n) number the leads and the second indices (Hi or
Lo) indicate the high and the lowpotential side of the quantum Hall resistance
device, respectively. Z
H
and R
H
with the single index H stand for the Hall impe
dance and the Hall resistance, respectively.
Each multiple lead has been represented as a Tnetwork: the lead impedance Z
k
is split into two equal halves (see Figure 6.42) consisting of the series resistance R
k
and the series inductance L
k
. Y
k
is the admittance of the lead consisting of the
capacitance C
k
with its associated loss factor tan d
k
. For simplicity, the indices Hi
or Lo have been omitted in (6.51) to (6.54).
~
Z
k
in (6.51) and (6.54) is called the
connecting impedance. The term (1 Z
k
Y
k
) in (6.54) is in any practical case equal
to 1 and the imaginary component of
~
Z
k
is usually small so that
~
Z
k
is approximately
equal to the resistance of the contact, R
Ck
, plus the series resistances of the parti
cular lead (see (6.54)). The terms Z
k
Y
k
/2 in (6.51) are independent of the number of
multiple leads, and they are identical with the usual quadratic cable correction
terms as they apply to any conventional twoterminalpair impedance.
2DEG
U = 0
(low)
B
Y
1Hi
Y
2Hi
I
1Hi
I
2Hi
I
2Hi
R
H
R
xxHi
R
xxLo
I
3Hi
R
H
I
3Lo
R
H
I
2Lo
R
H
R
H
R
C1Hi
R
C2Hi
R
C3Hi
R
C3Lo
R
C2Lo
R
C1Lo
I
1Lo Y
1Lo
Z
1Lo
I
3Lo
Z
3Lo
Z
2Lo
Y
3Lo
Y
2Lo
I
2Lo
k = 1
3
k = 1
3
U
(high)
Z
1Hi
Z
2Hi
Z
3Hi
Y
3Hi
I
3Hi
I
kHi
+ I
kLo
= 0
Figure 6.42 The equivalent circuit of a quantum Hall device with a tripleseries
connection scheme (n m 3) including the contact resistances and
the lead impedances and admittances. For simplicity, the outer
conductors of the coaxial leads are not shown
Impedance standards 175
The product terms in (6.51) are specific to the multipleseries connection
scheme. In the case of a twoterminalpair connection scheme (n m 1), (6.51) is
just the usual twoterminalpair formula for the resistance R
H
in series with the
connecting impedances
~
Z
k
. For every additional multipleseries lead, the potential
difference between each star point and the respective edge of the quantum Hall
device is reduced by an additional factor of
~
Z
k
/R
H
, as described by the product terms
in (6.51). If the number of leads is large enough and the connecting impedances are
small enough, the product terms become negligibly small. The cable correction now
corresponds to that of a fourterminalpair device (but with opposite sign), and, apart
from this cable effect, the Hall voltage between the two final potential contacts
becomes practically the same as the voltage at the two starpoint junctions.
To assess a particular connection scheme, the numerical values of the product
terms in (6.51) for a typical connecting impedance of a few ohms,
~
Z
k
/R
H
are a
few times 10
4
. Then the error of a doubleseries connection scheme is a few times
10
8
, which is not sufficiently small for precise AC quantum Hall resistance
measurements. Furthermore, as open leads are not allowed, only one potential
contact at each edge of the quantum Hall device can be provided with a lead so that
it is not possible to measure its longitudinal resistance. In contrast, triple and
quadrupleseries connection schemes sufficiently eliminate the effect of the lead
and contact resistances, and two or three potential terminals are available at each
side of the quantum Hall device for measuring longitudinal resistances.
Finally, a quantum Hall device with a multipleseries connection scheme (triple
series or higher) can provide a selfcontained combining network (see section 5.6). The
currents flowing through each defining Hall potential contact are automatically
adjusted to a value which, while not zero, is much smaller than what is needed for an
effective combining network like a current source or a Kelvin arm (see section 5.6.1).
Therefore, a quantumHall resistance device with a multipleseries connection scheme
can provide a selfcontained combining network. This simplifies both coaxial bridge
networks and their balancing procedures. Examples are discussed in section 9.3.
6.7.2 A device holder and coaxial leads
For AC quantum Hall resistance measurement, the quantum Hall device should be
mounted into a complete metallic shield to achieve a welldefined impedance
without any stray capacitive current to the cryostat (see Figure 6.43). This shield
cap
case
socket
coaxial lead
PCB
GaAs chip pin
Figure 6.43 A shielded device holder
176 Coaxial electrical circuits for interferencefree measurements
containing the quantum Hall resistance device is usually mounted at the lower end
of a probe that can be lowered to the centre of a superconducting magnet coil inside
a cryostat, as shown in Figure 6.44. A pumped
4
He system with a base temperature
of 1.5 K is sufficient for ordinary applications, but a
3
He system having a base
temperature of 0.3 K produces wider Hall plateaux and allows a wider scope for
research investigations. The coaxial multiple leads have to be fed through the probe
and the shield to the quantum Hall resistance device. Coaxial leads having a low
thermal conductivity minimise the thermal load on the cryosystem. The outer
conductors of the coaxial leads and the shield of the quantum Hall resistance device
50 cm
superinsulation
superconducting solenoid
inner vacuum chamber
1 K pot with liquid
4
He
sorption pump
gas cooled radiation baffles
4
He recovery
4
He gas
4
He liquid
3
He gas
coaxial leads from the QHE
device to the measuring circuit
top loaded probe with shielded
QHE device at its lower end
sliding seal
outer vacuum chamber
pot containing
liquid
3
He
holder with
QHE device
shutter valve
Figure 6.44 A device holder in a cryostat
Impedance standards 177
should be electrically isolated from the cryostat. If two shielded quantum Hall
resistance devices insulated and isolated from one another can be mounted sideby
side into the same cryomagnetic system, they can be measured in a quadrature
bridge (see section 9.3.7) to make a direct realisation of the SI farad.
To equalise the return currents in the outer conductors from the multipleseries
junctions outside the cryostat up to a point that is as close to the quantum Hall
device as possible, the coaxial leads should be fed through the shield of each
quantum Hall device. Their outer conductors as well as the device shield should be
connected to an internal star point close to the quantum Hall resistance device (see
Figure 6.43). If this is not done, the openloop inductance and capacitance of any
remaining length of unequalised inner conductors will cause a large unwanted
quadratic frequency dependence. The fourterminal resistance of the outerconductor
network is transformed by the equalisers to be in series with the quantum Hall
resistance, but it can be kept negligibly small with a starpoint connection. Its
fourterminal resistance can be measured with a fourwire microohm meter. If this
resistance is not negligibly small, a correction must be made for it.
The quantum Hall device is usually mounted on a printedcircuit board or a
ceramic carrier. Both types of carriers are provided with a certain number of pins
that can be plugged into the coaxial miniature sockets of the mounting system to
allow an easy exchange of quantum Hall devices. The printedcircuit board may
carry metal pads that act as gates or shields [34,38], and usually it is printed with
line conductors that reach from the pins to pads close to the quantum Hall device so
that connections can be made to the device with bonding wires. The bonding wires
should be as short as possible and the quantum Hall device should be fixed to the
printedcircuit board with glue to avoid ACdriven oscillations of the bonding wires
and the quantum Hall device in the magnetic field. As a further precaution, the
current contacts of the quantum Hall resistance device can be provided with two
parallel bonding wires of different lengths and therefore different resonance fre
quencies. The effect of a resonant frequency in one bonding wire will be shorted
out by the other. As there is capacitance between quantum Hall device and sur
rounding metals, the glue as well as the printedcircuit board material should have a
sufficiently low loss factor and a sufficiently high isolation resistance. If ungated
quantum Hall devices are used, the conductors on the printedcircuit board and the
star point in the case should not be too close to the quantum Hall device, otherwise
they act as unwanted gates. A standardised mounting system has been developed in
the frame of the EUROMET project No. 540 to allow the exchange of quantum
Hall devices between different national standard laboratories (see also Figure 6.34).
The coaxial leads require a minimum isolation resistance between the inner and
outer conductors. To obtain a desired relative uncertainty of 1 10
9
a moderate
isolation resistance of about 2 GW is sufficient for the defining leads of the coaxial
connections that have a typical inner conductor resistance of 2 W. For the other cables
even less will suffice. This is one of the useful advantages of a coaxial network over a
fourwire measurement that applies to AC as well as to DC measurements. In a four
wire measurement, the isolation resistances between the high and lowpotential line
conductors on the printedcircuit board as well as between the quantum Hall device
178 Coaxial electrical circuits for interferencefree measurements
and a splitgate or double shields are a shunt resistance in parallel to the quantum
Hall resistance. Therefore, these isolation resistances must be larger than 10 TW.
Isolation resistances can be checked with a DC teraohm meter.
The outer conductors of the coaxial leads and the shield of the quantum Hall
device must provide sufficient shielding to avoid any spurious crosstalk between
the inner conductors of different leads or between each inner conductor and any
other device, for example, the cryomagnetic system. The shielding can be tested
with a highresolution (&0.1 fF) capacitance bridge.
In general, any precise impedance standard requires a specific correction for
the impedance and admittance of the defining cables (section 5.4), and this also
applies to the quantum Hall resistance. To measure the cable parameters with a
precision LCR meter, two dummy printedcircuit boards without a quantum Hall
device should be used: one printedcircuit board with open terminals for the
measurement of the lead admittance, and a second printedcircuit board with short
circuited terminals for the measurement of the lead impedance. The cable para
meters should be measured in the cryomagnetic system under the same conditions
as for the actual quantum Hall resistance measurement.
The coaxial leads connected to the quantum Hall device are usually composed
of sections having a different impedance and admittance per meter (e.g. a section of
room temperature cable from the measuring bridge to the head of the cryostat and a
section of a thin cryogenic cable inside the cryostat) and the combination requires a
more complex cable correction. For defining cables consisting of two sections, the
cable correction is given in section 3.2.6. It is determined by the total lead series
impedance and shunt admittance if the different sections have the same char
acteristic impedance.
6.7.3 Active equalisers
AC quantum Hall resistance measurements usually require long coaxial leads for
the multipleseries connection scheme because the QHE occurs only at cryogenic
temperatures in a cryostat but the device is connected to measuring bridges at room
temperature. To keep the heat transport into the liquid helium bath of the device
small, the coaxial leads are usually thin, and thus the long and thin leads have
comparatively large resistances, typically about 2 W. This greatly reduces the
efficiency of the current equalisers involving these cables in their meshes and
increases the net current in these cables and its effect on the main balance. For
equalisers with the usual 1220 turns, the effect on the main balance is too large to
be accurately evaluated and corrected (as described in section 3.2.2). Active current
equalisers are recommended in connection with the ac quantum Hall resistance
because their lead capacitances are not as large as for passive current equalisers of
enhanced number of turns.
In a coaxial bridge network employing a quantum Hall device, it might not be
sufficient to use active equalisers only in the multiple leads of the device. For
example, the equalisers might be arranged in such a way that the two defining leads
of the quantum Hall resistance do not carry an equaliser. However, these two leads
Impedance standards 179
are part of other meshes in the bridge network and the corresponding equalisers
experience the large impedance of the long leads. This usually gives rise to sig
nificant net currents. If the resulting effect on the main balance (evaluated as
described in section 3.2.2) is too large, these equalisers should also be replaced by
active equalisers or by passive increasedefficiency equalisers. It is also possible to
arrange the equalisers in the bridge network in such a way that only one lead of the
quantum Hall device does not carry an equaliser. It then might be sufficient to use
active equalisers only in the multiple leads to the device but not in the rest of the
bridge network. However, the residual effect of all the equalisers in the network on
the main balance must still be evaluated.
6.7.4 Capacitive model of ungated and splitgated
quantum Hall devices
If no special precautions are taken, various internal and external capacitances draw
a current through a quantum Hall device, and this gives rise to additional Hall
voltages that contribute to the quantum Hall impedance Z
H
as well as to the long
itudinal impedance Z
xx
(see also section 6.6.2) [35]:
Z
H
R
H
n
wR
2
H
C
High
n
j tan d
High
n
C
Low
n
j tan d
Low
n
Z
ab
xx
R
ab
xx
n
wR
2
H
C
ab
n
j tan d
ab
n
_ _
The capacitances of the relevant edges as shown in Figure 6.41(a) and the
associated loss factors are labelled by the index n. We wish to discuss the situation in
general and so the meanings of C
1
, C
2
, etc. depend on the particular quantum Hall
resistance device and the metals, gates or shields in its surroundings. The contribu
tions to Z
H
originating from the high and lowpotential edge of the twodimensional
electron gas have opposite sign and the relevant edge regions reach from the parti
cular downstream defining potential contact to the lowcurrent terminal (Figure
6.41a). For simplicity, those additional correction terms of Z
H
that arise from the
multipleseries connection scheme (section 6.7.1) are not explicitly listed here again
but, of course, have to be added. The contributions to Z
xx
measured at two contacts a
and b originate only from the edge region between those contacts. R
xx
denotes the
AC longitudinal resistance that would be found if there were no capacitive effects,
and that in the kHz frequency range is practically equal to the DC longitudinal
resistance (i.e. for an ideal quantum Hall device it is zero in the plateau centre and
steeply increases at the plateau edges).
So far, three different contributions to the AC losses have been identified [37].
The main contribution stems from losses incurred by electrons in the two
dimensional electron gas caused by internal capacitances in it as well as by external
capacitances between it and surrounding conductors. The loss mechanism is
polarisation and this explains the linear frequency dependence of the measured Hall
resistance [39]. The loss factor increases linearly with current because both the
polarising electrical Hall field and the number of polarisable electrons in the
180 Coaxial electrical circuits for interferencefree measurements
twodimensional electron gas increase with current. Consequently, the AC losses of
the twodimensional electron gas appear as a linear frequency and current depen
dence of AC quantum Hall resistance measurements. Further, both the capacitance
and the associated loss factor measured as a function of the magnetic field increase
from the plateau centre towards the plateau edges, and this explains the curved
shape of the plateaux observed for both the quantum Hall and the longitudinal
impedances.
A second, much smaller, contribution has been attributed to lossy dielectric
materials located between the twodimensional electron gas and surrounding con
ductors. Relevant conductors could be the line conductors on a printedcircuit
board device carrier, gates or shields and lossy dielectric materials could be, for
example, the GaAs substrate and the device carrier.
The third small loss mechanism has been attributed to adsorbates on the sur
face above the twodimensional electron gas. In both cases, the relevant loss
mechanism is polarisation, which is approximately frequency independent, and
this, when combined with the prefactor w in the equations given above, results in a
linear frequency dependence of the quantum Hall and longitudinal impedances. As
the number of polarisable dipoles does not depend on the magnitude of the Hall
current, the corresponding contribution is current independent. The surface adsor
bates can be removed by standard vacuum outgassing techniques before the QHE
device is cooled down in the cryostat, and the other AC losses can be eliminated by
gates or shields, at least to within some limit as discussed below.
According to the current and frequency dependence of the individual loss
mechanisms, the current and frequency dependence of the quantum Hall and
longitudinal resistance can be empirically written as
ReZ
H
R
H
fk
0
fIc
0
higher order terms
ReZ
xx
R
xx
fk fIc:
where k and k
0
are currentindependent frequency coefficients and c and c
0
current
coefficients. This is the current and frequency dependence that has been empiri
cally known since the very first AC quantum Hall measurements but was not
understood for a long time.
6.7.5 Ungated quantum Hall devices
Conductors should not be located too close to ungated devices to avoid an unne
cessary increase of external capacitances beyond the inevitable capacitance
between the device and its shield. For shield dimensions much larger than the
dimensions of the device, the capacitance to the shield becomes independent of the
shield dimensions. Assuming an approximate relation k
0
/k &c
0
/c and a device that
satisfies the DC guidelines, the deviation of the AC quantum Hall resistance from
the quantised DC value is approximately proportional to the AC longitudinal
resistance. This proportionality can be measured and the AC quantum Hall resis
tance can be extrapolated to that which would be found if the longitudinal resis
tances were zero [40].
Impedance standards 181
6.7.6 Splitgated quantum Hall devices
Another approach is based on subjecting the device to electric fields from con
ductors near its edges. These conductors are termed as split gate, and can be
realised either as a split backgate where the device is mounted on two conducting
surfaces separated by a small gap along the centre line of a device (Figure 6.45) or
two sidegates [35,37].
split backgate
U
GHigh
= (1 + a)U
U
GLow
= bU
2DEG
GaAs substrate
Figure 6.45 A quantum Hall device mounted on a split backgate
Applying adjustable AC voltages of the same frequency as the measuring
frequency to each half gate generates an electric AC field that penetrates into the
dielectric 2DEG and biases the capacitances of the QHE device. This effect can be
described by two parameters l
High
and l
Low
:
Z
H
R
H
wR
2
H
C
High
1
_
j tan d
High
1
_
l
High
C
Low
1
j tan d
Low
1
_ _
l
Low
_ _
Z
High
xx
R
High
xx
wR
2
H
C
High
1
_
j tan d
High
1
_
l
High
Z
Low
xx
R
Low
xx
wR
2
H
C
Low
1
j tan d
Low
1
_ _
l
Low
Only the dominating internal capacitive effects are considered here. It is found
experimentally that the effect of a split gate on both Z
H
and Z
xx
can be described by
two multiplying factors rather than by sum terms and that each of these parameters
can be adjusted to be positive, negative or zero. If the component of the external
electric AC field in the plane of the 2DEG is equal to the internal Hall field, the
internal capacitances no longer load the quantum Hall device, that is, both para
meters l
High
and l
Low
are zero. Then, the main contribution to the frequency and
current dependence of the AC quantum Hall resistance and the AC longitudinal
resistances along both sides of the device vanish simultaneously. The criterion for
adjusting the gate voltages is the attainment of a zero AC longitudinal resistance at
both sides of the device (assuming again that the device satisfies the DC
guidelines).
182 Coaxial electrical circuits for interferencefree measurements
Apart from this special setting of the gate voltages, many other solutions exist
where the capacitive contributions to Z
H
at both sides of device are not individually
zero but mutually cancel each other. These solutions lie on a line in a two
dimensional gatevoltage diagram (and the special solution discussed above is a
point on that line. The criterion for adjusting the gate voltages is a zero current
coefficient of the AC quantum Hall resistance [34,35]. This approach is easier to
realise but it results in slightly curved Hall plateau if the capacitive effects at both
sides of the device have a different shape as a function of the magnetic field (an
effect that does not limit this approach), whereas the first criterion results in per
fectly flat plateaux (which is neither a necessary nor a sufficient criterion).
Both the extrapolation and the splitgate approach are only approximate solu
tions because residual frequency dependences of the order of a few parts in 10
8
/kHz
remain. In the case of the extrapolation method, the AC longitudinal resistance
shows a strictly linear current dependence whereas the current dependence of the
AC quantum Hall resistance shows a nonlinear saturation at higher currents.
Further, the internal and external capacitances contribute to the AC longitudinal
and quantum Hall resistance in a different way. Both effects limit this approach.
The splitgate approach is limited by the external capacitive AC losses in the GaAs
substrate that are not eliminated. However, the residual frequency dependences
of both methods are of the order of a few parts in 10
8
/kHz and are thus comparable
to the best calculable capacitors presently in operation. But, this is not the end of
the story.
6.7.7 Doubleshielded device
More than ten times better and, for all practical purposes, perfect results can be
obtained by a third approach. This approach considers where the relevant currents
flow and from this information, how to achieve that unwanted currents do not affect
the measurement. The strategy is not to eliminate the unwanted capacitances or to
use materials having lower loss factors but to handle all the currents in such a way
that the defining condition of the particular impedance standard is precisely met.
This is not only a simple and elegant strategy but also the most precise and sensible
solution of the problem.
The doubleshield approach is based on two conducting shielding boxes with a
narrow gap in between them. The device is located inside the boxes in such a way
that the defining Hall potential contacts are centred in the gap (Figure 6.46) [36,41].
I
I
U
high
U
low
= 0
high shield low shield
I
CH
I
CL xU
Figure 6.46 A doubleshield device
Impedance standards 183
The low shield is connected to the lowcurrent terminal so that the capacitive loss
currents are picked up and added to the measured current, whereas the shield
around the highcurrent terminal is connected to the outer conductor network so
that the capacitive stray currents from within it are intercepted and routed back to
the source so that they do not falsify the measurement.
As an additional detail, an adjustable potential xU has to be applied to the high
shield in order to electronically adjust the position of the shield gap relative to the
defining Hallpotential contacts and in this way to eliminate the residual AC losses
in the twodimensional electron gas due to an imperfect mounting of the device.
The criterion for adjusting the highshield potential is, exactly as for the splitgate
approach, the vanishing of the current coefficient of the AC quantum Hall resis
tance. This doubleshield approach also eliminates the effect of the AC losses of
adsorbates as well as the AC losses in the substrate that is the limitation of the split
gate approach. The purpose of the doubleshield method is to precisely meet the
defining conditions of an AC impedance standard, and this simple basic principle is
the reason of its success. Altogether, this approach allows the realisation of a
quantum standard of impedance that is more accurate than all conventional impe
dance artefacts. Residual parasitic capacitances can be measured using a scaledup
room temperature model of the device [41] and their effect on the phase angle
calculated so that the device constitutes a phase angle standard of as low an
uncertainty as can be obtained by other methods (see section 9.2.1).
To make use of such a precise quantum Hall impedance standard and to realise
a traceability chain like those shown in Figure 6.37 or Figure 6.38, we now need
appropriate coaxial bridges. Because a quantum Hall device can provide a self
contained combining network, the design of the bridges can be simplified. They are
described in sections 9.3.6 and 9.3.7.
References
1. Thompson A.M., Lampard D.G. A new theorem in electrostatics and its
application to calculable standards of capacitance. Nature. 1956;177:188
2. Mohr P.J., Taylor B.N., Newell D.B. 2008 CODATA recommended values of
the fundamental physical constants. Rev. Mod. Phys. 2006;80:633730
3. Lampard D.G., Cutkosky R.D. Some results on the crosscapacitances per unit
length of cylindrical threeterminal capacitors with thin dielectric films on their
electrodes. J. IEE. 1960;107c:11219
4. So E., Shields J.Q. Losses in electrode surface films in gas dielectric capaci
tors. IEEE Trans. Instrum. Meas. 1979;28:27984
5. Shields J.Q. Phase angle characteristics of crosscapacitors. IEEE Trans.
Instrum. Meas. 1972;21:36568
6. Rayner G.H. The timeconstant of carbon composition resistors. Br. J. Appl.
Phys. 1958;9:24042
7. Engheta N., Salandrino A., Alu A. Circuit elements at optical frequencies:
nanoinductors, nanocapacitors and nanoresistors. Phys. Rev. Letts. 2005;
95:095504
184 Coaxial electrical circuits for interferencefree measurements
8. Cutkosky R.D. Fourterminalpair networks as precision admittance and
impedance standards. IEEE Trans. Commun. Electron. 1964;80(70):1922
9. Solymar L., Shamonina E. Waves in Metamaterials. Oxford University Press,
Great Clarendon Street, Oxford, OX2 6DP, UK, 2009
10. Shields J.Q. Voltage dependence of precision air capacitors. J. Res. NBS.
1965;69c:26574
11. Kibble B.P., Rayner G.H. Coaxial AC Bridges. Bristol: Adam Hilger Ltd.;
1984. (Presently available from NPL, Teddington, TW11 0LW, U.K. www.
npl.co.uk)
12. Kibble B.P., Schurr J. A novel doubleshielding technique for AC quantum
Hall measurement. Metrologia. 2008;45(5):L2527
13. Campbell A. On a standard of mutual inductance. Proc. Roy. Soc. A.
1907;79(532):42835
14. Haddad R.J. A resistor calculable from DC to w 10
5
rad.s
1
. M.Sc. thesis,
School of Engineering and Applied Science, George Washington University,
1969
15. Grover F.W. Inductance Calculations: Working Formulas and Tables. Dover
Phoenix, 1973
16. Delahaye F., Goebel R. Evaluation of the frequency dependence of the
resistance and capacitance standards in the BIPM quadrature bridge. IEEE
Trans. Instrum. Meas. 2005;54(2):53337
17. Awan S.A., Kibble B.P. Towards accurate measurement of the frequency
dependence of capacitance and resistance standards up to 10 MHz. IEEE
Trans. Instrum. Meas. 2005;54(2):51620
18. Gibbings D.H.L. A design for resistors of calculable AC/DC resistance
ratio. Proc. IEE. 1963;110:33547
19. Bohacek J., Wood B.M. Octifilar resistors with calculable frequency
dependence. Metrologia. 2001;38(3):24148
20. Awan S.A., Kibble B.P. A universal geometry for calculable frequency
response coefficient of LCR standards and new 10 MHz resistance and
1.6 MHz quadrature bridge systems. IEEE Trans. Instrum. Meas.
2007;56(2):22125
21. Reilly S.P., Leach R.K., Cuenat A., Awan S.A., Lowe M. Overview of MEMS
Sensors and the Metrology Requirements for Their Manufacture. NPL report,
DEPCEM 008, section on MEMS Devices for Electrical Measurements,
2006, pp. 2531
22. Von Klitzing K., Dorda G., Pepper M. New method for highaccuracy
determination of the fine structure constant based on quantized Hall resis
tance. Phys. Rev. Lett. 1980;45:49497
23. Jeckelmann B., Jeanneret B. The quantum Hall effect as an electrical resis
tance standard. Rep. Prog. Phys. 2001;64:160355
24. Delahaye F. Technical guidelines for reliable measurements of the quantized
Hall resistance. Metrologia. 1989;26:6368
25. Delahaye F., Jeckelmann B. Revised technical guidelines for reliable DC
measurements of the quantized Hall resistance. Metrologia. 2003;40:21723
Impedance standards 185
26. Schurr J., Burkel V., Kibble B.P. Realizing the farad from two AC quantum
Hall resistances. Metrologia. 2009;46:61928
27. Ikushima K., Sakuma H., Yoshimura Y., Komiyama S., Ueda T., Hirakawa K.
THz imaging of cyclotron emission in quantum Hall conductors. Physica E.
2006;34:2226
28. Nachtwei G. Breakdown of the quantum Hall effect. Physica E. 1999;
4:79101
29. Ricketts B.W., Kemeny P.C. Quantum Hall effect devices as circuit ele
ments. J. Appl. D: Appl. Phys. 1988;21:483
30. Schurr J., Ahlers F.J., Hein G., Melcher J., Pierz K., Overney F., Wood B.M.
AC longitudinal and contact resistance measurements of quantum Hall
devices. Metrologia. 2006;43:16373
31. Melcher J., Warnecke P., Hanke R. Comparison of precision AC and DC
measurements with the quantised Hall resistance. IEEE Trans. Instrum.
Meas. 1993;42:29294
32. Delahaye F. Accurate AC measurements of the quantized Hall resistance
from 1 Hz to 1,6 kHz. Metrologia. 1994/95;31:36773
33. Delahaye F. Series and parallel connection of multiterminal quantum Hall
effect devices. J. Appl. Phys. 1993;73:791420
34. Delahaye F., Kibble B.P., Zarka A. Controlling AC losses in quantum Hall
effect devices. Metrologia. 2000;37(6):65970
35. Schurr J., Ahlers F.J., Hein G., Pierz K. The AC quantum Hall effect as a
primary standard of impedance. Metrologia. 2007;44(1):1523
36. Kibble B.P., Schurr J. A novel doubleshielding technique for AC quantum
Hall measurement. Metrologia. 2008;45(5):L2527
37. Schurr J., Kibble B.P., Hein G., Pierz K. Controlling losses with gates and
shields to perfect a quantum Hall impedance standard. IEEE Trans. Instrum.
Meas. 2008;58:97377
38. Overney F., Jeanneret B., Jeckelmann B. Effects of metallic gates on AC
measurements of the quantum Hall resistance. IEEE Trans. Instrum. Meas.
2003;52:57478
39. Jeanneret B., Overney F. Phenomenological model for frequencyrelated
dissipation in the quantized Hall resistance. IEEE Trans. Instrum. Meas.
2007;56:43134
40. Overney F., Jeanneret B., Jeckelmann B., Wood B.M., Schurr J. The quan
tized Hall resistance: towards a primary standard of impedance. Metrologia.
2006;43:40913
41. Schurr J., Kucera J., Pierz K., Kibble B.P. The quantum Hall impedance
standard. Metrologia. 2011;48:4757
186 Coaxial electrical circuits for interferencefree measurements
Chapter 7
Transformers
Transformers couple circuits without the need for any direct conducting connection
between them, although often a single connection is added to make the potential of
one winding definite with respect to another. The magnetic flux caused by the
current in one winding, usually designated the primary winding, threads the other
windings, and the aim of a good transformer design is to make this flux coupling as
complete as possible. This is usually achieved by linking the windings with a high
permeability magnetic core to provide an easy path for the magnetic flux.
In instrument transformers, we describe the cores as toroids wound from
extremely highpermeability material, formed into a very thin ribbon to avoid flux
exclusion by eddy currents at frequencies up to 1 MHz. To enhance the com
pleteness of flux coupling further, the windings are usually tightly coupled, for
example, being the separate strands of a tightly spun rope appropriately connected
in series or parallel.
Enclosing the primary winding around its toroidal core in a toroidal shield, as
shown in Figure 5.4 ensures that the entire magnetic flux of the primary winding
threads equally, with extremely high accuracy of parts in 10
9
, two or more sec
ondary windings. The ratios of the voltages induced in secondary windings are then
quite accurately equal to the ratio of the number of turns of these windings.
Transformers can be divided into two categories. In one, the purpose is to
transmit electrical power from one circuit to another, commonly at a different
voltage. Unless they are autotransformers having only a single tapped winding,
they may also provide isolation of one network from another so that there is no
current flow between the networks. The design of isolating transformers is such that
they transmit power efficiently, solely by the magnetic coupling, and a high pre
cision of voltage or current ratio between the primary winding and the secondary
windings is not an important consideration.
In this chapter, we are concerned with the other category of transformer that may
deliver some power, but the design objective is rather to approach an ideal trans
former. In an ideal transformer the emf induced in each and every turn is equal and the
voltage developed by a winding is very close to the sum of these induced emfs, i.e. to
one part in a million or better. Isolation may also be required. In the case of current
transformers, the ratio of the currents in the windings is the important quantity, and
the accompanying voltages are only of incidental interest.
Sometimes the two windings are, in fact, continuous and a tapping point is
provided, as in an autotransformer, and sometimes a continuous winding is divided
into ten or more sections by nine or more taps.
Other uses for transformers in measuring and sensing circuits include impe
dance matching and noise matching (see section 2.1.3).
7.1 General considerations
The easiest and most accurate way to compare impedance standards is in terms of
the voltage or current ratio of transformer windings, because the ratio between
output taps of a properly constructed transformer is not affected, at least to parts
in 10
8
, by aging or moderate changes in its environmental conditions (position,
temperature, pressure, humidity, etc.). Moreover, if an accuracy of a part in a
million is satisfactory, the ratio of the transformer can often be taken to be equal to
the ratio of the number of turns of the windings. When lower uncertainty is
required, a method of calibrating the transformer is needed.
We first discuss some general considerations that limit the performance of
transformers, then we describe how to construct and design transformers to mini
mise these effects, and finally, we discuss calibration methods. This final part can
be omitted on a first reading.
7.1.1 The causes of departure from an ideal transformer
It is often convenient when analysing the role of a transformer in a circuit to con
struct its equivalent circuit from a combination of an ideal transformer and circuit
elements, which represent the imperfections in the characteristics of the actual
transformer. In an ideal transformer, (1) there are no energy losses either in the
windings or in the core; (2) the only magnetic flux is that which threads equally all
the turns of the ratio windings; (3) flux from a current in any of the ratio windings
threads completely the other ratio windings and (4) there are no significant
capacitances within or between the ratio windings or to the magnetic core or any
surrounding screens.
In the following account of various transformers and the circuits in which they
are employed, we shall be considering the causes of departure from an ideal trans
former and how they can be minimised and their effects eliminated in some circuits.
The departures of a transformer from ideal characteristics arise from several
causes.
The magnetic material of the core absorbs, as well as stores, energy when it is
magnetised; there is a magnetic power loss, and because magnetisation character
istics are nonlinear, the power loss cannot be strictly represented by current
flowing through a simple resistance. Eddy currents induced in the core material
give rise to additional power loss.
The windings of the transformer possess resistance, which, at higher fre
quencies, will be increased by skineffect eddy currents both within the conductors
and as induced in adjacent conductors. There will also be capacitance between
188 Coaxial electrical circuits for interferencefree measurements
turns and between layers of a winding; it is shown later that these distributed
capacitances can be represented by a single lumped capacitance between the ends
of a winding. In addition, some small proportion of the flux linking one winding
will follow a path that does not link another winding. This small amount is usually
called leakage flux and is the main cause of leakage inductance, which affects the
ratio of loaded ratio windings. The design and construction of transformers used for
sensing and measurement purposes aim at minimising the effects of some or all of
these imperfections.
The core material should have a high stored magnetic energy compared with
the energy lost per cycle. Within the restrictions imposed by other considerations,
the wire for the windings should be thick to minimise the resistance and self
inductance of the wire itself; the latter contributes to the leakage inductance. The
capacitances within a winding can be reduced by spacing the individual turns and
layers, but the advantages gained are offset by increased leakage flux and higher
winding resistances resulting from the use of finer wire for the windings. In addi
tion, it may be necessary, for example, to consider the dielectric loss of the wire
insulation and other quantities of secondary importance.
Although a transformer may depart considerably from the ideal, it will still
constitute a precise and stable ratio device provided that the causes of departure do
not change. One cause of variability is the magnetic state of the core; this is con
sidered in the next section.
7.1.2 The magnetic core
Until recently, the core material was always one of the mumetal or permalloy class
of nickeliron alloys having a composition of 75% Ni and 25% Fe; the alloy with
the highest permeability is known as super mumetal in the United Kingdom or
supermalloy in the United States and is used where its higher cost is not an
important factor. Like all magnetic alloys, its magnetic properties are far from
linear. The relative permeability of typical mumetal alloys is shown in the graph of
Figure 7.1. It increases by a factor of about 4 from its initial value at very low flux
densities up to a maximum that can be as high as 200 000 at a flux density of 0.2 T.
A solid magnetic core can be used only if the magnetic flux does not change,
because when magnetic flux in the core changes, an emf is induced in the core in
planes perpendicular to the direction of the flux, and in accordance with Lenzs
law, eddy currents are set up in the core such that the flux changes are opposed
(Figure 7.2a). The emfs and hence the eddy currents are proportional to the rate of
change of magnetic flux. This effect occurs in any piece of metal in a varying
magnetic flux, but the effect is enhanced in magnetic materials by their higher
permeability. For a given uniform flux density B, the only way of reducing the
magnetic flux
_
BdS is to reduce dS, where S is the cross section perpendicular to B.
This is usually done by building up a magnetic core from thin sheets or stampings
or turns of a ribbon (Figure 7.2b) insulated from one another, the sheets being in the
direction of the flux. Since the induced emf is d
_
BdS/dt, the larger the rate of
change, the smaller the
_
dS required to reduce the induced emfs and eddy currents.
Transformers 189
That is, higher frequencies require thinner sheets. The Vitroperm and Ultraperm
materials, from which the more recent cores are made, are created as amorphous
(i.e. noncrystalline or glassy) ribbons only a few micrometres thick by extremely
rapid cooling of the molten material on a cold rotating cylinder. The thinness of the
ribbon, together with the high resistivity of the material, ensures even greater
freedom from eddy current effects than the older mumetal cores. Cores wound
from the thinnest amorphous metal tapes are usable up to megahertz frequencies.
Unfortunately, the space factor (the proportion of the total cross section occupied
by magnetic material) becomes worse for thinner ribbons as the necessary insula
tion between adjacent turns of ribbon becomes a larger proportion of the total
core volume.
If the flux density is raised appreciably above that which produces maximum
permeability, the material saturates, and at the peaks of a large applied sinusoidal
0
0.1 0.2
10 10
4
5 10
4
magnetic induction/T
r
e
l
a
t
i
v
e
p
e
r
m
e
a
b
i
l
i
t
y
Figure 7.1 The permeability of magnetic core material as a function of flux
density
f
(a) (b)
f
Figure 7.2 Eddy currents induced by flux changes in (a) a solid magnetic core
and (b) a ribbonwound core
190 Coaxial electrical circuits for interferencefree measurements
waveform, the permeability is very low. As a result, the input impedance of a
winding falls at these excitation peaks so that it is no longer much larger than the
output impedance of a practical source. Excessive voltage drop occurs in the source,
the waveform is clipped, and the proportion of harmonics generated in the current
and voltage waveforms associated with the transformer is increased considerably.
The permeability and the power loss at a given flux density change with time
and depend on the past history of the material. A DC current in a winding or
mechanical and magnetic shocks are also likely to change the permeability and
increase the loss and flux leakage between windings, altering the precise voltage
ratios of the transformer, but the magnetic properties of a transformer core of this
material can always be restored to a fairly welldefined state when required by
demagnetising the material. A transformer core can be demagnetised by an alter
nating current in a winding whose peak value is large enough to thoroughly saturate
the core and then smoothly decreasing the current to zero in a time long compared
to its period. Using a much lower frequency than the normal operating frequency
of the transformer will lower the reactance of the core and therefore the driving
voltage needed to produce a saturating current. Unfortunately, following mechanical
shock or demagnetisation, the permeability decreases with time, and this spontaneous
change renders the demagnetised condition somewhat uncertain.
Eddy currents may be regarded as reducing the effective permeability of the
magnetic core and increasing the total power lost via the electrical resistance of the
current path during the alternating magnetisation.
Toroidal cores are universally used in precise ratio transformers because of
their symmetry. However, this symmetry is not quite perfect because they are made
by winding a ribbon of material as shown in Figure 7.3. The ribbon has a beginning
on the inside of the toroid and an end on the outside, and a small proportion of the
alternating flux, instead of being confined to the interior of the toroid, will jump
from one end of the ribbon to the other, as illustrated in the figure. Also, another
stray flux
high permeability
ribbon
Figure 7.3 Flux leakage from a ribbonwound toroidal magnetic core
Transformers 191
small proportion loops from outside to inside of the core on both sides of it. This
looping flux is greater near the inner and outer ends of the spiral strip. Effects
arising from these stray fluxes can be, for all practical purposes, eliminated by
enclosing the core and the winding energising it in a toroidal screen of mumetal.
7.1.3 The windings; the effect of leakage inductances, capacitances
and resistances
We now consider these three effects, all of which may be to some degree unde
sirable in a particular device and whose individual minimisation are unfortunately,
in practice, incompatible.
If the conductors comprising two windings could be brought into exact geo
metrical coincidence in space, all the flux that links one winding also links the
other. Since it is impossible to get two conductors to spatially coincide, a small
proportion of the flux generated by a current flowing in one winding will fail to
thread the other winding. This small proportion causes a small inductance, termed
the leakage inductance, of the windings. This leakage inductance manifests itself as
an internal inductance L of the windings, which alters their voltage ratio when
current is drawn from one or more of the windings by a load.
Current I drawn by a load Z causes a flux f, of which a small amount df does
not thread the second winding, as shown in Figure 7.4.
U
1
U
2
I
load
df
f
Z
Figure 7.4 The effect of leakage flux on a transformer
df gives rise to a missing emfd(df)/dt L
1
dI/dt and modifies the ratio U
1
/U
2
by this amount, so L
1
is a real physical inductance in series with the winding. The
resistance of the winding and the loss of L
1
can be represented as a lumped resistor
R
1
in series with the winding, so that we have the equivalent circuit of Figure 7.5
where the windings are ideal, that is, perfectly coupled and of zero resistance.
In a transformer where the windings are wound very closely together to
minimise this effect and spaced a distance away from the core or magnetic screens
by insulation materials, etc., the presence or absence of the core or magnetic
192 Coaxial electrical circuits for interferencefree measurements
screens will not greatly affect the magnitude of the leakage inductance because the
flux in the core or screens threads both windings. Also the leakage inductance and
capacitance of windings do not depend much on the core, so that it is again a useful
approximation in considering them to think of the device as aircored. Any mag
netically permeable material interposed between the windings for whatever purpose
will, of course, greatly increase the leakage inductance.
The apparent resistance of a winding is also not affected much by the presence
of the core, so that it is nearly the same as if the winding were air cored. The low
frequency and DC resistance value will be augmented at higher frequencies by
skin and proximityeffect eddy currents within the conductor, particularly if the
winding is closely packed or multilayered.
In a voltage ratio transformer, the potential of successive turns of the windings
increases in a regular manner until the full output voltage is reached across the
complete winding. Between all pairs of turns, capacitance currents will flow as
shown in Figure 7.6.
etc.
U
Figure 7.6 The internal capacitances of a transformer winding
The net effect of these distributed capacitances can be represented by a single lumped
capacitance. Each component capacitance dC, by virtue of transformer action, is
equivalent to a capacitance dCr
2
across the winding, where r is the fraction of the
voltage across dC compared with the total voltage across the winding. Alternatively,
a string of capacitances in series are equivalent to a single capacitance dC/n across
the winding. These relations are valid for all frequencies for which the phase of the
L
1
R
1
L
2
R
2
Figure 7.5 The equivalent circuit of a transformer having leakage flux
and winding resistances
Transformers 193
current throughout the winding is constant; in practice, this even applies to fre
quencies well above selfresonance.
Since L and R also arise from distributed phenomena, they must be shown on
the winding side of this distributed shunt capacitance. Again, the winding itself is
now assumed to be free of selfcapacitance effects. The equivalent circuit is as
shown in Figure 7.7.
The current I drawn by the selfcapacitance alters the output voltage U of the
winding to U dU. As
I % jwCU
dU IR jwL jwCUR jwL
w
2
LC jwCRU
The real part of dU is negative, and hence, the inphase component of voltage
is increased either by the presence of selfcapacitance or by an external capacitive
load connected across the output terminals. L, R and C are best regarded as para
meters describing the behaviour of the winding.
Cutkosky [1] has pointed out that the deliberate introduction of a small
impedance z in common between the two windings, as shown in Figure 7.8, can be
advantageous. The output impedance of one section can be decreased at the
expense of another, and a better overall compromise can be reached. The current
R L
U U dU
C
I
dU
Figure 7.7 The equivalent circuit of a transformer winding including the effects of
leakage inductance L, resistance R and internal capacitance C
Z
1
Z
2
U
2
U
1
C
D
A
B
z
Figure 7.8 Connecting a small resistance between the windings
194 Coaxial electrical circuits for interferencefree measurements
drawn by a load between terminals A and B flows through z, thus altering the
potential of terminal C, so that the ratio of U
1
/U
2
is immune to this loading if
z Z
1
/(U
1
/U
2
1). In the case of a 10:1 ratio, z Z
1
/9.
7.1.4 Representation of a nonideal transformer: the effect of
loading on its ratio windings
There are two kinds of transformer used in bridge networks. These two kinds may
formally be termed two and threewinding transformers. Twowinding transfor
mers are usually known as inductive voltage dividers (IVDs) or their dual, induc
tive current dividers, and are discussed in sections 7.3.1 and 7.3.2. Threewinding
transformers, usually known as voltage (or current) ratio transformers, are dis
cussed in sections 7.3.77.3.11. The essential difference between these two kinds is
that the energising current from a source flows through some or all of the windings
of a twowinding transformer, causing small potential differences across their
leakage inductances and winding resistances, but in a threewinding transformer,
the energising current flows in a separate primary winding. The inherent accuracy
of the latter arrangement is therefore greater.
An equivalent circuit of a twowinding transformer is drawn in Figure 7.9.
Z
M
Z
1
n
1
n
2
Z
2
U
2
U
1
Y
a
Y
b
Figure 7.9 Equivalent circuit of a twowinding transformer
A source may be connected across either winding, or if the windings have a
common connection as shown, it may be connected across both. The winding and
load Z drawn to the immediate right of the core do not represent an actual winding
and load. They represent the extra impedance resulting from loss mechanisms
presented to the driven winding, which might be the one with n
1
turns or the one
with n
2
turns or both in series. Z
1
and Z
2
represent the series resistances and leakage
inductances, and Y
a
and Y
b
, the internal shunt admittances of the windings. Anal
ysis, neglecting product and higher terms, of the small quantities z
1,2
yields
U
1
U
2
r 1
Z
1
rZ
2
1 rZ
Y
b
rY
a
Z
1
r
2
Z
2
r1 r
_ _
7:1
Transformers 195
where r n
1
/n
2
. Equation (7.1) shows how Z
1
, Z
2
, Y
a
and Y
b
cause the voltage ratio
U
1
/U
2
to depart from the nominal ratio of the winding turns.
The equivalent circuit of a threewinding transformer (which has a separate
energising winding), when measuring a ratio of admittances Y
1
/Y
2
, is shown in
Figure 7.10. It shows the effect that loading produces on the transformer ratio.
M
D
Z
1
n
1
U
n
2
U
Z
2
U
2
U
1
Y
1
Y
2
Y
a
Y
b
Figure 7.10 The equivalent circuit of a threewinding transformer
If the flux induced in the core by the primary winding in turn induces an emf U
per turn in the secondary windings, the ratio of emfs U
1
and U
2
applied to the
admittances Y
1
and Y
2
to be compared is
U
1
U
2
n
1
U
n
2
U
_ _
1 Z
2
Y
2
Y
b
1 Z
1
Y
1
Y
a
Y
2
Y
1
7:2
Equation (7.2) shows that the ratio of the voltages applied to the admittances
Y
1
and Y
2
is the turns ratio n
1
/n
2
modified by the (usually small) product terms
Z
1
(Y
1
Y
a
) and Z
2
(Y
2
Y
b
).
Z
1
and Z
2
are typically 10 mH in series with 0.1 W, so that when comparing
admittances Y
1
and Y
2
of 10
3
S at an angular frequency of 10
4
rad/s, the correc
tions to the ratio U
1
/U
2
contained in the square brackets of (7.2) are of the order of
10
4
and must be accounted for, or otherwise eliminated, in accurate work.
Simply connecting the energising source across Y
1
and Y
2
will eliminate their
effect, since the current through them is drawn from the source rather than the
transformer. The effect of Y
a
and Y
b
is not entirely eliminated, as they represented
the distributed admittances of the windings.
The error in the voltage ratio of a transformer caused by the loading of the
shunt admittances Y
1
Y
a
and Y
2
Y
b
presented by the measuring circuit to its
output ports can be readily found and accounted for by measuring this shunt
admittance, increasing it temporarily by a known ratio with another shunt admit
tance and extrapolating back to the zero load condition. The internal impedances
Z
1
and Z
2
of the transformer winding, which cause the loading errors, can also be
calculated using (7.2).
196 Coaxial electrical circuits for interferencefree measurements
7.1.5 The twostage principle
If an only approximately correct flux in a core has been created by a primary or
magnetising winding, the deficiency can be made up by a much smaller flux in a
second core. This latter flux is produced by a winding energised by the emf caused by
the discrepancy between the exact flux required and the smaller flux provided by the
magnetising winding. As applied to voltage transformers, whose equivalent circuit is
shown in Figure 7.11, it is desired that the total flux should correspond exactly to the
source emf U
1
defined at the input terminals, but in the ordinary singlestage design,
the flux corresponds to an emf smaller by an amount dU
1
because of the voltage drop
the primary current suffers in flowing through the impedance associated with the
primary windings. This impedance is composed of the resistance r of the primary
winding plus the leakage inductance l between primary and secondary windings.
R L
dU
U
2
U
1
n
1
n
2
I
mag
I
2
= 0
U
1
dU
Figure 7.11 The equivalent circuit of the primary of a voltage transformer
Hence, this emf U
1
dU is equal to the rate of change of flux in the trans
former core.
dU I
mag
R jwL
And the emf appearing across the output terminals is
U
2
U
1
dUn
2
n
1
instead of the ideal value U
2
U
1
n
2
/n
1
.
A second core added and wound as shown in Figure 7.12 is driven by the lost
emf dU, which, to a good approximation, can be transformed to dUn
2
/n
1
and added
R L
dU
I
2
U
1
I
mag
n
1
n
2
n
1
n
1
n
2
(U
1
dU)n
2
/n
1
U
1
n
2
/n
1
dUn
2
/n
1
Figure 7.12 A second core added to a voltage transformer
Transformers 197
to the emf in the secondary winding to give the corrected voltage
U
1
dUn
2
n
1
dUn
2
n
1
which equals U
1
n
2
/n
1
, as was required.
An approximate analysis of the device can be made as follows, with reference
to Figure 7.13.
I
2
dU
Z
2
Z
3
Z
SC
I
mag
Z
1
Z
2
Z
3
Z
SC
U
1
Z
1
Z
1
Z
P
Z
2
Z
3
Z
SC
where Z
P
and Z
SC
are the impedances of the wound first and second cores, and
Z
1
. . . , Z
5
are composed of the resistances and leakage inductances of the five
windings.
U
2
U
1
dUn
2
n
1
dU I
2
Z
2
Z
3
n
2
n
1
U
1
1
Z
1
Z
2
Z
3
Z
1
Z
P
Z
2
Z
3
Z
SC
_ _
n
2
n
1
7:3
I
2
is small because of the high impedance Z
SC
of the lossy inductance of the
second core.
Hence, since Z
1
( Z
P
and Z
2
Z
3
( Z
SC
, the actual voltage ratio U
2
/U
1
differs from the turns ratio n
2
/n
1
by approximately the product of two terms, Z
1
/Z
P
and (Z
2
Z
3
)/Z
SC
.
In practice, a separate auxiliary transformer can be avoided by the construction
shown in Figure 7.14. Remember that a winding threads all cores and their
U = I
mag
Z
1
U
1
U
2
Z
1
Z
2
Z
sc
Z
3
Z
p
Z
4
Z
5
I
2
I
mag
n
1
n
2
n
1
n
1
n
2
(U
1
dU)n
2
/n
1
Figure 7.13 Parasitic impedances of a transformer having a second core
and winding
198 Coaxial electrical circuits for interferencefree measurements
windings drawn to the left of it, so that winding 1 is wound only around core 1 and
windings 2 and 3 are wound around both cores 1 and 2. By virtue of being wound
around core 1, winding 2 senses the flux f
1
and behaves like the second winding on
the main transformer of the previous scheme, and by virtue of being wound around
core 2, it can generate flux f
2
in core 2. Since the emfs in winding 2 from f
1
and f
2
are additive, we can represent the device, as in the previous figure, as two windings
in series, each with its individual series impedance Z
3
and Z
2
.
Similarly, winding 3 is around both cores, and it can be represented as the
secondary windings of both main and auxiliary transformers in series, each with its
individual impedance Z
4
and Z
5
.
Examples of this general concept of two staging will be discussed in the
descriptions of particular devices later in this chapter.
The voltage across a second core of a voltage ratio transformer is in general
higher by a few per cent than the voltage across the firststage winding. This is
because of the selfresonance induced by the selfcapacitance of the winding in
conjunction with its large inductance. The flux caused in the second core by the
selfresonant current is easily opposed by an external adjustable current source
connected across the winding, and in this way, the overall voltage of the potential
windings, which are threaded by this flux, can be altered without altering their
voltage ratio (see section 7.3.8).
7.1.6 Electrical screens between windings
The capacitive current between two windings can be reduced to zero by providing
two screens between them, one screen being connected to each winding. Both
screens are arranged to be at the same potential but are not directly connected
within the transformer, as shown in Figure 1.19. As in section 1.1.2, the current
through any residual capacitance between the screens can be diverted through a
lowimpedance single conductor so that it does not affect nonisolated sensitive
circuitry (e.g. a phasesensitive detector) in the measurement network supplied by
the transformer. Addition of this conductor creates a mesh, which needs to be
equalised by an equaliser in the input cable to the sensitive circuitry.
The screens must have overlapped breaks in them so that they do not produce a
shorted turn on the transformer. The correct way to arrange the breaks is illustrated
in Figure 1.27.
f
1
f
2
1 2 3
n
1
n
1
n
2
U
2
U
1
Figure 7.14 Avoiding the need for a separate auxiliary transformer
Transformers 199
7.2 Constructional techniques
7.2.1 Design of transformer windings
A toroidal core for a transformer imposes some practical constructional limitations
on the windings. The simple windings, which can be put on by a toroidal winding
machine, do not usually produce an optimum transformer design, and the best
compromise for a given purpose will usually necessitate winding by hand.
In this section, we describe techniques that help to ensure that a toroidal
magnetic field is created by the energising winding.
Any transformer winding should be uniformly spread around the toroid so that
its two ends are adjacent, as shown in Figure 7.15. Such a winding will be rea
sonably unaffected by external flux f in its plane because the flux will divide
approximately equally between the two sides of the toroid and in doing so will
induce equal and opposite cancelling voltages in the two halves of the winding.
In progressing around the toroid, however, the winding forms a loop in the
plane of the toroid of a size equal to its mean diameter (Figure 7.16). The loop area
may be decreased by carrying the conductor from the end of the winding round the
toroid in the opposite direction to the winding progression so that it finishes adja
cent to the beginning of the winding. This is called an antiprogression turn.
is
equivalent
to
f
Figure 7.16 A toroidal winding progressing around a core
f
Figure 7.15 An uniformly wound toroid
200 Coaxial electrical circuits for interferencefree measurements
A small disadvantage of an antiprogression turn as sketched in Figure 7.17 is
that it does not follow the mean path of the toroid. Therefore, flux in the toroidal
plane may still couple into the winding, or an energised transformer core may
produce flux in this plane. Thus, an antiprogression turn is only an approximation
to a perfect solution to the problem because it evidently does not occupy the same
physical volume as the actual winding, and therefore, some flux can still thread the
combination.
A better solution is to use the AyrtonPerry or bootlace technique as sketched
in Figure 7.18. Half the number of turns required are put on the toroid in the usual
way, and the rest are wound on top of them, winding in the same sense through the
core but in the opposite direction around it until the beginning of the winding is
reached.
A still more complex winding in which the progression of application of the
turns is as indicated in Figure 7.19 has the additional advantage of decreasing the
selfcapacitance of the winding considerably.
is
equivalent
to
f
Figure 7.17 A transformer winding with an antiprogression turn
Figure 7.18 A bootlace technique for transformer winding
Transformers 201
finish start
Figure 7.19 Progression of a bootlace winding to minimise selfcapacitance
Windings of the kinds we have described, which do not have significant emfs
induced in them by external fluxes, are termed astatic.
In some of the following sections, the pictorial illustrations of various wind
ings are not shown with an astatic feature for reasons of clarity. They should be
understood to be astatically wound in practice.
The circuit shown in Figure 7.20 is a rapid way of counting a large number of
turns wound on a core. A small number of turns m (typically 10), which can be
quickly and reliably counted, is temporarily wound and connected in seriesaiding
(i.e. they thread the core in the same sense) with the n turns to be counted. Because
of topology, n and m are exact integers corresponding to the number of passages of
the conductor through the hole of the toroid.
The windings are connected to a multidecade IVD (see section 7.3.1) via a
detector as shown, and a generator is connected across the whole circuit. The circuit
n
x
calibrating
IVD
x = m/n
U
m
D
U = 0
Figure 7.20 A circuit for counting the number of turns of a winding on a core
202 Coaxial electrical circuits for interferencefree measurements
is not coaxial, but no great care needs be taken about stray capacitances because
this circuit of backtoback transformers has a lowimpedance and because the detector
is positioned in the lowpotential side of the circuit. Also, because the greatest part of
the magnetic flux associated with the circuit is within the wound core, the open
meshes of this modestaccuracy circuit need only be minimised by sensible con
ductor routing.
The dial settings x of the IVD are adjusted until the inphase component of the
detector is nulled. The quadrature component caused by parasitic capacitances is
small and irrelevant. Then n m/x, and n can be calculated. The result will not be
exactly an integer because of the finite resolution of the IVD and inaccuracies of the
circuit, but it can be verified by repeating the procedure with a different value of m.
A simpler technique, which is adequate if the number of turns, or the turns
ratio, is not too extreme, is to simply measure the ratio of the voltages with an
AC voltmeter, as shown in Figure 7.21.
n
m
U
n
/U
m
= n/m
D
U
Figure 7.21 The ratio of numbers of turns measured using an AC voltmeter
7.2.2 Techniques for minimising the effect of leakage inductance,
winding resistance and the capacitances of ratio windings
In this section, we describe winding techniques that optimise the performance of
ratio windings.
It is first necessary to recall the concept of leakage inductance (section 7.1.3).
This small inductance arises from flux caused by current flowing in one section of a
ratio winding, which does not thread the other sections. In a rope construction,
described in sections 7.2.4 and 7.2.5, each strand of the rope is a section of the
winding. Hence, the output impedance of a winding has an inductive component
arising from these leakage fluxes, as well as each winding possessing resistance. To
attain a voltage or current ratio as nearly equal to the turns ratio as possible,
capacitance between sections of the windings should be minimised.
Transformers 203
The requirements are unfortunately mutually exclusive; minimum leakage
inductance requires that the separate ratio windings are as close together as possi
ble, turn by turn, to one another, and this inevitably increases the capacitance
between sections. The inphase ratio error of a voltage or current ratio transformer
is affected by the load the distributed capacitance offers to the impedance of the
leakage inductance. That is, the effect is proportional to the product of the leakage
inductance and the capacitance, so that if a winding design increases one and
decreases the other, the ratio error tends to remain much the same.
It is usually desirable however to minimise the effect of output impedance on
winding ratios to make the influence of external shunt capacitive loading of cables,
etc. as small as possible. This means that leakage inductance should be decreased at
the expense of interstrand capacitance, and for the quadrature component, the
resistance of the windings should be as small as possible. This is achieved in var
ious ways by the winding designs described in sections 7.2.37.2.5.
If the windings are made as individual successive single layers on a toroid with
interposed layers of insulation, the selfcapacitance of each layer of winding will be
small and the capacitance between layers will be reduced. The remaining capaci
tive currents can be further reduced by arranging that those parts of the windings at
similar potentials are adjacent. The effect of interwinding capacitance can be
reduced to zero by appropriate screens placed between windings, as noted in
section 1.1.5.
If high accuracy of the ratio of voltage windings and small leakage inductance
are required, it is sometimes permissible to allow the capacitances between two
windings to increase, but to make the principal capacitances symmetrical, so that
they do not affect the ratio. This is achieved in the ordered rope winding descri
bed in section 7.2.5.
7.2.3 Bifilar winding
The simplest possible windings are the bifilar windings of a 1:1 ratio transformer.
If the winding is constructed symmetrically, the resulting 1:1 voltage ratio is
usually closer to the 1:1 ratio of the turns than for the winding ratio of other
transformers. To achieve this, a pair of conductors, using wire from the same reel to
ensure equality of resistance and insulation properties, is twisted together so tightly
that they lie in contact throughout their length, as shown in Figure 7.22a. This
ensures symmetry of capacitance, resistance and leakage inductance along their
length, so that the induced voltages are affected identically and their ratio is
accurate. Because the windings are twisted, any small differences of emf induced in
the pair by nontoroidal stray fluxes will sum to zero to a very close degree over the
full length of the winding.
Figure 7.22b has been derived from the pictorial representation of Figure 7.22a
by conceptually slitting the core in the vicinity of the zero tap and opening it out to
a straight line. We can now imagine the windings unravelled by pulling the +U
point up and the U point down so that the zero points of the windings coincide, as
shown in Figure 7.23.
204 Coaxial electrical circuits for interferencefree measurements
0
C
3
C
1
C
1+
C
3+
C
2
C
2+
(a) (b)
+U
U
0
+U
U
Figure 7.23 (a) Schematic representation of a bifilar winding. (b) Capacitances
between the conductors of a bifilar winding
Near the +U and the U ends of the windings, equal capacitive loads C
1+
and
C
1
to zero exist. Further capacitances symmetrical about zero occur throughout
the winding, so that the distributed capacitive loadings on the +U and U sections
are the same. For example, for C
2+
C
2
, the voltage changes due to them are the
same and the ratio of the output voltages remains accurately +1:1.
We have assumed that a rope consisting of the twisted pair of conductors is
wound on the toroid, so that capacitances between turns of the rope are small
compared to those between the twisted conductors. Twisting will also have an
averaging effect on the small betweenturns capacitances. Moreover, they are
largest between those adjacent turns that have the smallest potential differences, so
that the capacitive current is small. Also, symmetry ensures that capacitances are
equal for the two windings; that is, C
3+
C
3
, etc.
It is, therefore, not surprising that bifilar wound transformers can readily be
made with +1:1 ratio errors of the order of 1 in 10
9
at audio frequencies where
core permeabilities are high. Their ratio errors remain within 1 in 10
6
up to 1 MHz,
provided cores suited to these higher frequencies are used.
7.2.4 Rope winding having randomly arranged strands
A less perfect solution, but one that yields other voltage ratios and is easier to
implement, is to take the conductors that are to form the ratio windings, lay them
+U
+U
U
U
0
0
(a) (b)
Figure 7.22 A bifilar winding: (a) pictorial representation and (b) diagrammatic
representation
Transformers 205
into a tight parallel bundle, twist one end to wind the complete set into a rope and
wind the complete rope a sufficient number of times (10100) around the core, as
sketched in Figure 7.24. At the ends of the rope the conductors are then joined and
can be connected to switches or connectors to form the ratio taps of the completed
transformer.
+U
+U
U
U
0
0
+2U
+10U
+10U
+2U
(a) (b)
Figure 7.24 (a) A rope winding on a transformer core. (b) A schematic
representation
Thus, a rope of 11 strands can be connected to form a +10:1 ratio transformer.
For the same reasons as for binary winding, the conductors should be as close
together as possible, and twisted, but not as in an ordinary rope where preferred
strands lie in the centre and others around the outside. Rather, the position of the
strands in the rope should be randomised by periodic interweaving before twisting so
that interwinding capacitances and stray flux coupling are also randomised.
When the individual strands are connected in series to form a tapped winding,
pairs of strands with relatively large interstrand capacitances should be connected
so that there is only a small voltage between them.
The departure from the nominal values of ratios equal to the ratios of numbers
of turns produced in this way is typically a few parts in 10
7
at frequencies around
1 kHz. So long as the conductors are prevented from physically altering their
position in the completed transformer and the magnetic state of the core is kept
reasonably constant, the ratios are stable and can be calibrated with an uncertainty
of the order of 1 in 10
9
(see section 7.4).
7.2.5 Ordered rope winding
An ordered rope winding is the very opposite of a random rope winding. The
sections of voltage division are conductors spun around a central core, which can
either be a plastic of low dielectric constant or another insulated conductor of
greater cross section, as shown in Figure 7.25.
206 Coaxial electrical circuits for interferencefree measurements
(a)
(b)
1
2
3
4
5
6
7 8
9
10
1
Figure 7.25 (a) An ordered rope of conductors for a transformer ratio winding.
(b) A cross section
When this construction is achieved by spinning the strands of conductors as a
single layer around a dielectric core of appropriate diameter, the capacitances
between the adjacent strands are equal. Their symmetry ensures that they have little
effect on the voltage ratios of the winding. The capacitance between the first and last
strands could be merely a load on the generator connected across the winding as a
whole or could be reduced by interposing a dummy strand of low dielectric constant
plastic. Other capacitances between nonadjacent strands are much smaller. Because
leakage inductances are also quite small, this winding technique is an appropriate
compromise if the ratio errors are required to be small up to higher frequencies.
The technique can be employed to achieve two separate advantages. When
employed in an IVD (with the addition of a dummy strand), the interstrand capa
citances forma capacitive divider in parallel with the inductive voltage division. Since
the capacitances are approximately equal, their effect on voltage division is small, and
the device retains good accuracy up to higher frequencies of the order of 100 kHz.
For use in a 10:1 voltage ratio transformer, the rope is spun around a con
ductor of greater cross section. The leakage fluxes between the spun 010 sections
and the central conductor, which is the 0 to 1 section, are very small. A trans
former made with such a rope can therefore be expected to exhibit small output
leakage inductances and to have a very stable ratio. The interstrand capacitances
are now higher and not evenly distributed, being predominately between the 0 to 1
central conductor and the other sections, and so the transformer will give best
performance at lower frequencies, below about 10 kHz.
If good performance is needed above 10 kHz, it might be better to wind an IVD
with an ordered 11strand rope (with a dummy strand) and to permanently connect the
end of the first section where it connects to the second section as a zerooutput tap.
7.2.6 Magnetic and electric screens
It is desirable to confine the flux produced by a winding to the interior of a toroidal
space bounded by the outer surface of a winding. Therefore the winding should
Transformers 207
approximate a uniform current sheet, but a practical winding consists of a limited
number of turns of wire of finite cross section, which allow flux leakage. Non
toroidal flux will also result from core imperfections, as described in section 7.1.2.
Both the effects of stray flux and capacitive currents can be reduced to negli
gible proportions by toroidal conducting screens of highpermeability mumetal.
Where a demountable joint is needed, it must be provided with a large overlap so
that the reluctance of the air gap is small enough. Soft soldering can be used to give
electrical continuity. Figure 1.10b shows a section through a toroidal shield; con
nections to the wound toroid in the annular space enter through the small extension
on the left. An insulating sleeve prevents a disastrous complete conducting turn
threading the core.
Shields can also be formed as washers and annular strips from a bendable form
of mumetal, which can be cut with scissors; although this is not as efficient as
seamwelded and heattreated thick sheet material, the resulting shield can be
sufficiently good to be useful.
Screens made of thick copper have also been used. These rely for their mag
netic effect on the skin depth phenomenon; eddy currents induced in them tend to
prevent the passage of unwanted flux. The skin depth at 1 kHz is about 2 mm, so
this technique is only useful at higher frequencies.
Sometimes the unwanted emergence of magnetic flux is not an important
effect in the device concerned, and a purely electrical screen having an overlapping
insulated gap suffices. This can be made of any conducting material, and con
ducting paint applied to a layer of an insulating tape is often employed. Whatever
the material used, thought must be given to the possible threading of the core by the
capacitive current flowing to and within the screens.
Consider a screen around a core energised with a winding, as shown in cross
section in Figure 1.10a. Because the screen is a single turn around the energised
core, there will be a corresponding potential difference between its ends, and a
capacitive current will flow between them and through the screen. Therefore, a
small, but finite, current threads the core, and its effect may be significant in the
performance of the complete device. In the case of a magnetic screen, the amount
of overlap is a compromise between a nearperfect magnetic screening path and the
large capacitance of a big overlap. This capacitance permits a capacitive current to
flow, driven by the emf induced in a single turn of the screen.
7.2.7 Testing the attainment of a nearly toroidal field
We have emphasised that a major objective when constructing accurate transfor
mers is to ensure as far as possible a toroidal magnetic flux by means of windings
properly distributed around the core and by using surrounding toroidal magnetic
screens where necessary. Then even if the magnetic state of the core is affected by
partial magnetisation or magnetic shock, the total flux still threads all ratio wind
ings external to the toroid, and their ratio accuracy is unaffected.
It is useful to be able to test whether a nearly toroidal field has been attained,
and a test is easily carried out by winding two equal coils, A and B, in opposing
208 Coaxial electrical circuits for interferencefree measurements
senses and connecting them to a detector as shown in Figure 7.26. T is the wound,
energised toroid being tested, which may also have a surrounding toroidal magnetic
screen.
A B
T
D
Figure 7.26 Testing the perfection of a nearly toroidal field
First, the coils A and B are brought together as closely as possible. Coil A is
kept fixed with respect to the toroid, whilst coil B is moved around to various parts
of the circumference. If the excited toroid assembly has a totally toroidal flux, no
voltage will be registered by the detector, and if the detector does register a voltage,
its ratio to that induced in a single turn of coil A is a measure of the imperfection.
7.2.8 Connections to the output ports
The basic winding techniques for the secondary windings have already been
described, but the method of bringing out taps to the coaxial connectors requires
some consideration if the highest possible accuracy is to be attained.
A good design is to mount the coaxial output connectors and the connector to
the primary winding on one side of the enclosing magnetic box screen, but insu
lated from it. Thick wires from the outer conductor of the ratio winding connectors
are routed alongside the tap connections to a point well within the volume of the
box where they are joined together and to the zero tap as illustrated in Figure 7.27.
In this way, unwanted small mutual inductances between the connectors to the taps
and selfinductances from only an inner conductor to a connector passing through a
hole in a magnetic sheet are avoided. The purpose of this method of mounting the
output sockets is to ensure that the potentials of the outer conductors are all well
defined and small and are not affected by the flow of current returning to the
transformer through the outer conductors of other taps. This would happen if a
multiplicity of other paths were provided by mounting the outer connectors directly
and uninsulated on a conducting panel. The voltage of every tap is well defined,
being that between the inner and outer conductor of its coaxial output socket. The
connectors to the first (magnetising)stage winding are also insulated from the box
and their outer conductors connected together, as shown.
Transformers 209
coaxial
output sockets
insulated
from box
wire making
a single
connection
to the box
magnetising stage
Figure 7.27 Connections to the output ports of a transformer
7.3 Types of transformers
The foregoing general principles have been applied to the construction of several
kinds of transformers which have been found useful in accurate impedance
comparing networks.
7.3.1 Inductive voltage dividers
An IVD is an autotransformer in that it has no separate primary winding. A single,
tapped winding fulfils the roles of both primary and secondary windings. It usually has
several decades of output subdivision to provide a fine resolution of its voltage ratio.
These devices are of the greatest utility as they provide a voltage whose
magnitude is related to the source voltage by an accurate and stable ratio, which can
be finely adjusted as required and whose phase angle, if little current is drawn,
departs from that of the source by as little as 1 mrad.
An ordered rope winding (see section 7.2.5) is best for all frequencies; for low
and intermediate frequencies and the later decades of subdivision, a random rope
winding is reasonable.
Since we are presupposing a construction in which all the flux threads all the
windings, the voltages across the turns intervals n
1
and n
2
of Figure 7.28 must be
such that U
2
/U
in
n
2
/n
1
. Hence, U
2
can be altered in finite steps by switching in
different turns to alter the ratio of integers n
2
/n
1
. Multioutputs from paralleled
switching are possible, and often two independent voltage outputs are provided in
this way.
As there is a practical limit to the number of turns of reasonably sized wire that
can be used (10
2
10
3
is usual), the fineness of the steps, which is set by the ratio of
numbers of complete turns, is insufficient for most applications. Hence, some
method of achieving what is, in effect, a fraction of a complete turn is required.
210 Coaxial electrical circuits for interferencefree measurements
Suppose another winding of m
1
turns is wound on the core and connected to a
winding of m
2
turns on the core of a second transformer as shown in Figure 7.29.
m
1
n
1 n
3
n
2
U
3
U
2
U
out
U
in
m
2
f
1
f
2
Figure 7.29 Further subdivision of the input voltage of an IVD
Only a small current flows because of the large inductance of the m
2
turns
on the second transformer; and the flux linkages in cores 1 and 2 are almost equal
that is, m
1
f
1
m
2
f
2
. If n
3
turns are wound also on core 2, the emf induced in these
turns is, to a good approximation,
U
3
n
3
df
2
dt
n
3
m
1
m
2
_ _
df
1
dt
n
3
n
2
_ _
m
1
m
2
_ _
U
2
since n
2
df
1
/dt U
2
and U
2
(n
2
/n
1
)U
in
The winding of n
3
turns can be put in series with the winding of n
2
turns to
achieve a total output voltage
U
out
n
2
n
1
n
3
m
1
n
1
m
2
_ _
U
in
7:4
The procedure can be iterated to further cores to provide further subdivision as
required.
U
in
U
out
n
1
n
2
Figure 7.28 A simple IVD
Transformers 211
Various modifications of this basic concept are employed to simplify
construction or improve accuracy.
(i) The n
3
turns may be a tapped portion of the m
2
turns. Switch selection of taps
results in the circuit illustrated in Figure 7.30.
U
0
0.1
0.2
0.3
0.4
0.6
0.7
0.8
0.9
1.0
0
0.01
0.02
0.03
0.04
0.06
0.08
0.09
0.1
0.05
etc.
U
out
= 0.57...U
Figure 7.30 Switched selection of the taps of an IVD
(ii) Economical use of cores can be achieved by combining each successive pair
of decades onto one core, that is, putting ten sections of 10p turns with taps
between the sections on the core together with ten sections of p turns simi
larly tapped. The circuit is then as shown in Figure 7.31.
U
0
0.1
0.2
0.3
0.4
0.6
0.7
0.8
0.9
1.0
0
0.01
0.02
0.03
0.04
0.06
0.05
0.08
0.09
0.1
0
0.001
0.007
0.003
0.004
0.006
0.008
0.009
0.01
0.005
0
0.0001
0.0002
0.0003
0.0004
0.0006
0.0007
0.0009
0.001
0.0005
U
out
= 0.5728U
Figure 7.31 Minimising the number of cores needed for an IVD
212 Coaxial electrical circuits for interferencefree measurements
An IVDis mostly used as a stable ACpotential or current divider, for which a low
output impedance is not of the utmost importance. Hence, a rope method of con
struction with a winding technique that limits the response to external flux is suitable.
The first decade of the above example would consist of a rope of ten strands wound 10p
times on the core with a bootlace or nonprogressive winding; the ends of appropriate
strands are connected together at the fixed contacts of the firstdecade switch.
Either the ends of the strands can be labelled before the strands are twisted up
into a rope, or opposite ends of the same strand can be identified with an AC
continuity tester after the rope has been twisted up and wound on the core. (A DC
continuity tester could leave the core in a magnetically saturated condition even if it
supplied only a fraction of an ampere turn.) Sometimes an 11th strand is connected
across the nextdecade switch (as shown by the dotted wire in Figure 7.30); we will
describe its purpose later.
The second decade of Figure 7.31 is astatically wound with p turns of an
11strand rope, 10 strands of which are connected to the second switch in a similar
fashion to the connections to the first switch. The 11th strand connects across the
10q turns of the 10 or 11strand rope of the next transformer assembly, which also
has q turns of a 10strand rope. The switches are connected as shown to form a
complete fourdecade divider. The construction technique can be iterated in an
obvious fashion by adding further twodecade cores to make a divider of 2n dec
ades. Four, six and eightdecade dividers are commercially produced in this way.
The input impedance of the device may be increased by increasing p up to the
point where the comparatively large selfcapacitance of the rope winding resonates
with its inductance at the frequency wat which the divider is to be used. Atypical value
would be p 5 at 1.6 kHz (w 10
4
). Above the resonance frequency, the input
impedance decreases again. The first core, therefore, has a large ratio of m
r
A/l, where A
is the crosssectional area; l, the mean magnetic path length; and m
r
, the relative per
meability of the core material, to maximise the inductance for a given p. Also, if B
max
is
the maximum flux in the core for which m
r
is still reasonably high (just below the flux
at which the core material saturates), B
max
A must be sufficient enough so that
100pAwB
max
exceeds the highest source voltage to be applied to the divider. Typically,
if A 10
4
m
2
, B
max
0.2 T, w 10
4
rad/s, p 5 and U
max
100 V. If minimum
quadrature component of the output voltage is desired rather than high input impe
dance, it is better to use the minimumnumber of turns required to support the intended
voltage rating. The input impedance of the ratio winding may be increased in another
way, by the twostaged technique described in section 7.1.5. No interwinding screens
are necessary in these devices, and the internal capacitive loads alter the output voltage
by typically less than one part per million of the input voltage at moderate frequencies.
Some insulating spacing between the ropes forming successive decades on the same
core is sufficient to reduce the error in the output voltage to 0.1 ppmor less of the input.
The transient conditions that occur as the switches are operated can be trou
blesome. It is preferable that the switches are of the makebeforebreak type, that
is, contact is made with the next position before it is broken on the present position.
One section of the winding is thus momentarily shorted, imposing a transient low
impedance load on the source but not greatly affecting the output voltage. This is in
Transformers 213
contrast to the other possible situation of breakbeforemake switching, where
the circuit to the succeeding core is momentarily open and the collapse of flux in
the core produces a large momentary emf at the output.
A better arrangement can be made and is worthwhile for the first decades
where these transient effects are greatest. By means of twopole switching with
makebeforebreak contacts on one pole and breakbeforemake contacts on the
other, first, a resistor of the order of 10 W can be switched across the taps, then the
output contact is moved, and finally, the resistor is switched out.
In the design described above, each succeeding core is energised by a separate
winding, which is a strand of the p turns per section. A similar arrangement can be
made for the second winding on a core; although it might seem unnecessary as the
second winding is already correctly energised by virtue of threading the same core as
the first winding, it is advantageous for two reasons. First, particularly on the first
core of a divider, the magnetising current flows through the windings comprising the
first decade of division but in the absence of the extra connected strand, not through the
seconddecade windings. The voltage drop caused by this current traversing the leak
age inductance and winding resistance then causes a lack of equality of potential across
one section of the first decade to that across the whole of the second decade. This
situation is alleviated by connecting an extra strand of the firstwinding rope across the
entire second winding. Second, the leakage inductance that exists between the first
decade rope and the seconddecade rope is greatly reduced by coupling them together
in this way. This results in a reduced output impedance of the complete divider.
If we consider now the connection between core assemblies, then the second
assembly will draw its magnetising current from the first. The resistance and leakage
inductance of the energised winding on the second core will cause a loss of emf across
it because the magnetising current must flow through this impedance. Therefore, the
total voltage across a succeeding core assembly will not equal the voltage difference
between taps on the winding preceding it. Thus, small unwanted discontinuities will
exist in the output voltage of the complete divider as a function of switch settings.
These discontinuities can be overcome, although the accuracy is not improved,
by the switching shown in Figure 7.32, which arranges that succeeding windings
are energised directly by the emf between taps.
Sometimes when using IVDs as adjustable voltage sources if their output is
significantly loaded, the changes in output impedance as the ratio switches select
various sections of the windings cause discontinuities. This is particularly trouble
some when attempting fine adjustment using the lowest dial settings because there
can be gaps in the range of output voltages. This defect is overcome by the circuit of
Figure 7.32b, where the two decades of finest subdivision of output voltage are
obtained from a 100:1 stepdown transformer supplied by separate switches from the
core two decades nearer the input. The impedance of the windings and their varia
tions supplying the input to this transformer is thereby reduced by (100)
2
, and this,
together with the small and constant impedance of the transformer secondary wind
ing, greatly reduces the output impedance of the last two decades.
The switch positions show a selected output voltage of 0.372026 times the
input voltage.
214 Coaxial electrical circuits for interferencefree measurements
0
10
11
U
2 1 3 4
U
out
= 0.3725 U
U
out
= 0.372026 U
0
10
11
U
1:100
5 6 2 1 3 4
0
10
11
U
U
out 1:100
1
2
3
4
5
6
n
10n 10m
m
10k
k
(a)
(b)
(c)
Figure 7.32 (a) Twotap switching of an IVD. (b) The two final decades are via a
stepdown transformer. (c) Reducing the number of cores
Transformers 215
As before, by winding a rope of 10p turns and a rope of p turns, adjacent pairs
of cores can, in fact, be combined as one core with a saving of space and magnetic
material, as in Figure 7.32c.
We shall use the formal representations of an IVD drawn in Figure 7.33a.
Sometimes a single device containing some common circuitry, which provides two
independent outputs, will be convenient, and this will be represented as shown in
Figure 7.33b.
The twostage principle can provide further improvement in accuracy, as
described in the next section.
7.3.2 Twostaged IVDs
The internal impedances of a singledecade IVD winding can be represented as
shown in Figure 7.34.
The series impedances Z
q
, Z
q
0
, etc. arise from the resistance and leakage
inductance of the individual intertap windings (e.g. the individual strands in a rope
construction). They cause an additional contribution i
mag
n
q1
Z
q
Z
0
q
to the emf
of the nth tap.
The input impedance of the device at the generator terminals is Z jwL R,
where L and R represent the inductance and loss of the wound core. Z is much
larger than the impedances Z
q
and Z
q
0
so that I
mag
is approximately U/Z, where U is
the source emf. wL and R are likely to be about the same size.
Typically, for 50 10 turns of 10
3
m diameter copper wire wound on a core
of 10
4
m
2
crosssectional area and 0.2m mean magnetic path, wL would be about
3 10
4
W, and the resistance and leakage reactance of the winding would be about
0.1 j10
2
W at w 10
4
rad/s. The wire resistance will vary by 1% or more
between strands, and the leakage inductance by the order of 100%. Several of the
Z
q
are involved in intermediate tap voltages, so that the output voltage could be in
error in magnitude and phase by up to a part in a million of the input voltage.
The twostage approach can effect an immediate improvement by providing a
ratio winding with a very low value of I
mag
. The circuit for a single decade is drawn
in Figure 7.35.
(a) (b)
Figure 7.33 Formal representation of (a) an IVD having a single output and
(b) an IVD having two independent outputs
216 Coaxial electrical circuits for interferencefree measurements
I
mag1
I
mag2
U
out
U
in
Figure 7.35 The twostage principle applied to an IVD
etc.
I
mag
= U/(R + jwL)
U
in
U
out
Z'
4
Z'
3
Z'
1
Z'
0
Z
0
Z
1
Z
4
Z
2
Z
Figure 7.34 The internal impedances of a singledecade IVD
Transformers 217
The inphase error in the output voltage at low or moderate frequencies
(40400 Hz) can be as low as a few parts in 10
9
, but the quadrature voltage may be
ten or more times larger.
Further decades of voltage division can be added by providing more windings
and cores as before, with succeeding stages also having separate magnetising
windings and two cores. Figure 7.36 is an example of a possible circuit.
The less elaborate singlestage construction would be used for further decades
after the first, second and possibly third, the relative loss in accuracy being by then
insignificant. Again, with ratio windings having 10p turns and p turns, two
successive decades can be combined on one core assembly needing only one
magnetising winding.
Leakage inductance and interstrand capacitance will be the same for a one or
twostaged transformer, and so a twostaged transformer will have similar high
frequency imperfections as a singlestage transformer.
7.3.3 Injection and detection transformers
In designing coaxial bridges, it is very useful to be able either to introduce a gen
erator of a small additional voltage DU at a point along the inner of a coaxial cable
(Figure 4.11a) or to detect the vanishing of a current I at a point along the inner
(Figure 4.11b).
The duality principle suggests that both of these objectives can be met using
the same device, which is either a coaxial injection or a detection transformer,
depending on which of the roles it is fulfilling.
10
0
decade 1
0
1
10
decade 2
11
(a)
(b)
last decade penultimate decade
U
in
U
out
Figure 7.36 (a) A manydecade twostaged IVD. (b) Its formal representation
218 Coaxial electrical circuits for interferencefree measurements
Figure 7.37 A pictorial section and a formal representation of an injection or
detection transformer
A cutaway diagram (Figure 7.37) will make the construction clear.
A hollow cylindrical conducting can has coaxial connectors fitted axially to its
ends. Inside the can, the connectors are joined by a shielded conductor. The shield
is not continuous but has a short gap at the centre. By adding nearly touching metal
discs to the shield on either side of the gap so that the gap is much deeper than its
width, the capacitance between the inner conductor through this gap to objects in
the annular interior of the can is made extremely small. Alternatively, the shield
can have a short insulated overlapping section to serve the same purpose. A high
permeability toroid, bootlace wound with a large number of turns (typically 100),
surrounds the coaxial line, and the ends of the winding are brought out to two
coaxial connectors on the can. If only a single input or a detector connection is
required, the other socket is shorted.
The gap at the centre of the shield locates the exact point along the inner
conductor at which the current can be detected to be zero. Figure 7.38 shows this is
so because the current through the distributed cable capacitances on either side of
the gap flows through these capacitances and back through the outer conductor.
I
I
I
I
Figure 7.38 Capacitive currents on either side of the shield surrounding the inner
conductor
This construction is a n:1 transformer whose primary is the n turns on the
toroid and whose secondary is the single turn of the axial conductor. The gapped
screen and the can form a complete electric screen, which isolates the primary from
the secondary. The axial symmetry of the construction ensures that the device
neither emits nor responds to external magnetic flux.
Transformers 219
A construction in which the primary winding is isolated is also possible, as
shown in Figure 7.39. Use of this device in a network will remove the need for one
current equaliser.
Figure 7.39 An injection/detection transformer having an isolated winding and its
formal representation
A detector transformer, which responds to the total current in a coaxial cable,
that is, to any imbalance in the supposedly equal and opposite currents in the inner
and outer conductors, is also required. Figure 7.40 illustrates a suitable construction
in which a highpermeability toroid is wound with a hundred turns, which begin
and end on the inner and outer, respectively of a coaxial socket. The whole winding
and connections to the socket are enclosed in a toroidal shield of highpermeability
material, which is connected to the outer of the socket. The shield has an over
lapped annular gap so that it does not constitute a shorting turn.
A cable to be tested for the presence of unbalanced current is threaded through
the toroidal assembly, for which it is a singleturn primary winding, and any net
current in the cable causes an emf to appear at the coaxial detector socket. The
component can also fulfil the dual purpose of injecting a current into the low
impedance circuit of the outer conductors in a mesh.
7.3.4 Use of an injection transformer as a small voltage source
If a voltage generator of output U is applied to one input of the nturn secondary
(typically n 100) and the other coaxial socket is shorted, a voltage of U/n is
generated in the inner conductor at the centre of the device.
B
D
to detector
~
I'
I
U
det
= (I I' )
S
B dS
S
d
dt
Figure 7.40 Across section through a detection transformer, which detects or injects
an unbalanced current in a coaxial cable and its formal representation
220 Coaxial electrical circuits for interferencefree measurements
By virtue of the superposition theorem, the use of the device can be extended
to generate independent inphase and quadrature voltages in the inner conductor.
The inphase generated voltage U
1
and the output U
2
of the quadrature phase
shifting circuit are arranged to have low output impedances compared with the
impedance of the 100turn primary of the injection transformer which is typically
10
4
W, and are applied one to each of the two ends of the primary winding. The
schematic representation of the device is then as shown in Figure 7.41a. The
practical circuit in Figure 7.41b could have R
S
10 W and C
S
0.1 mF for an
angular frequency w 10
4
rad/s.
U
1
U
1 U
2
U
2
C
S
R
S
(a) (b)
Figure 7.41 An injection transformer as a small voltage source
A typical figure for the output impedance of the generator and phaseshifting
network is 10 W. The generator at either input to the primary is connected to the
high impedance of the primary in series with the small output impedance of the
other generator so that the phase of either generator is preserved and an emf equal
to 10
2
(U
1
jU
2
) is generated in the coaxial inner secondary with an injection
impedance of only (10
2
)
2
10 10
3
W.
If the inductance of the primary is insufficiently large compared with the
output impedance of the generator, the twostage technique (see section 7.3.2) may
be employed. The circuit is then as shown in Figure 7.42. A twostaged device
contains two toroidal cores; the first is wound with a first stage or magnetising
U
1
'
U
1
U
2
'
U
2
Figure 7.42 A twostaged voltage injection transformer
Transformers 221
winding. The second toroid is placed alongside the first, and the second stage is wound
threading both cores with the same number of turns as in the first stage. The pairs of
coaxial inputs terminating these windings should be fed by two tracking inphase and
two tracking quadrature sources. The sources feeding the firststage winding are cor
rected by the much smaller currents in the secondstage winding. The accuracy of the
emf generated in the inner can be improved by this twostaged technique from the
order of 1 in 10
3
to inphase or quadrature ratio errors of 1 in 10
6
or better.
7.3.5 Use of an injection transformer as a detector of zero current
A current flowing down the central conductor threads the core and induces magnetic
flux in it. This in turn induces an emf in the manyturned secondary, which can be
sensed by a detector plugged into one of the output sockets, the other being shorted.
The precise point along the inner conductor at which the condition of zero
current is sensed is defined by the position of the narrow gap in the outer, since up
to this point some current can flow down the inner and across to the outer through
the capacitance between them (Figure 7.38). In an equalised network, this current
must return along the outer in the same direction as both sides of the gap are at the
same potential, because they are shorted by the case of the device around the out
side of the toroid, and no capacitive current will flow across the gap.
When this device is used merely to detect the presence or absence of a current,
the accuracy of its ratio is unimportant, and a singlestage construction is all that is
necessary. The factor of n
2
in the transformed impedance, where n is the number of
turns on the secondary, can be very valuable when noise matching a detector
having a high input impedance to a circuit of low impedance (see section 2.1.3).
7.3.6 Calibration of injection transformers and their associated
phase change circuits
The inphase calibration of an injection transformer A can be carried out with
reference to an IVD of adequate accuracy. A circuit for a singlestage device is
shown in Figure 7.43. A twostaged device would need a separate feed from a
tracking IVD to supply its magnetising winding.
D
A
Figure 7.43 Calibrating the inphase component of a voltage injection
transformer
222 Coaxial electrical circuits for interferencefree measurements
A null of detector D indicates zero current threading its associated detector
transformer, which in turn indicates equality of potential between the IVD output
and that of the injection transformer A so that the IVD reading for this condition
calibrates the ratio of A.
A 90
phase change circuit and the injection transformer used with it are best
thought of as a single entity to be calibrated together.
For this calibration, we need components of known value and phase angle, and
the best components that are readily available are gasfilled or primarystandard
quality fusedsilicadielectric capacitors, which have phase angles of the order of
10
6
, and small commercial electronic resistors of about 100 W value. The self
inductance of these resistors is approximately balanced by their selfcapacitance as
commonly constructed, and phase angles of less than 10
5
up to frequencies of
10
4
rad/s are usual (see section 6.4.11).
These components can be assembled into a bridge to calibrate a quadrature
injection network as shown in Figure 7.44a.
As can be seen from Figure 7.44b, the bridge is balanced and the detector
nulled if the current through the capacitor equals that through the resistor, that is,
a bwCR. The quadrature generator circuit is shown within broken lines.
Component values of about R 100 W and C 100 pF whose exact values are
known are suitable for calibrating a phase change network of 10
2
attenuation for
use with a 10
2
ratio injection transformer at a frequency of 10
4
rad/s.
In this discussion, only the source of a quadrature voltage and its injection have
been considered. It is a straightforward matter to test the complete injection system
with its two IVDs, one for the inphase voltage and one for the quadrature. The
former should be set to zero when checking the quadrature voltage and vice versa.
D
R
C
bU
bU
jaU
jaU
C
S
R
S
U
D
R
C
I
(a) (b)
Figure 7.44 Calibrating a voltage injection network. (a) Coaxial circuit.
(b) Principle of the circuit
Transformers 223
In most applications where a voltage is to be introduced into a network, the
accuracy of the magnitude of the inphase voltage is more important than that of the
quadrature voltage. It is desirable, therefore, that any inphase component introduced
by the quadrature injection network should be trimmed to be as small as possible.
A very convenient, simple and stable trim component for the 90
phase change
circuit associated with an injection transformer is made by putting a few turns of
insulated copper wire wound up into a circular coil to form an inductor of value of
the order of 100 mH in series with the resistor. The value of the resistor is so chosen
that the resistance of the combination is just less than the nominal value required.
The impedance of the combination must be much less than the input impedance of
the injection transformer (about 10
4
W). R should therefore be 10 W or less. The
phase of the injection thus generated is adjusted to 90
10
i1
a
i
_ _
10U 1 0:1
10
i1
a
i
_ _
The 10:1 voltage ratio of the transformer is
10U 1 0:1
10
i1
a
i
_ _
:U
which departs from the nominal 10:1 ratio by the fraction
0:1
10
i1
a
i
7:5
In the singleconductor practical circuit shown in Figure 7.51, the output vol
tage of the calibration transformer is assumed to be nominally equal to the intertap
voltage of the transformer to be calibrated. The adjustable series voltage aU is
supplied by an injection transformer, phaseshift network and dual output IVD as
described in section 7.3.4. We have supposed for simplicity that the detector is an
isolated, batterypowered type.
This simple bootstrap approach works well enough to moderate accuracies at
low frequencies but suffers from four serious drawbacks:
(i) If the overall potential of the surroundings is that of the zero tap, capacitive
currents will flow from both transformers through all parts of the calibration
loop to the surroundings, and some of these currents will flow through the
detector and give incorrect balances.
(ii) As drawn, the circuit is that of a singleconductor, openloop construction
instead of a conductorpair coaxial design. Consequently there will be many
mutual inductive interactions between the loop and other parts of the circuit,
again causing wrong balances.
230 Coaxial electrical circuits for interferencefree measurements
(iii) The use of an isolated detector may place a limit on the achievable sensi
tivity if its characteristics are not suited to the lowimpedance circuit of the
loop.
(iv) The relative constancy of the output voltages of the two transformers
depends on the constancy of both of their primarytosecondary voltage
ratios. These cannot be expected to be as stable as the voltage ratio of, for
example, two secondary windings of one transformer because of the inevi
table lack of spatial coincidence of the primary and secondary windings
owing to the necessary interposing of shields. The amount of stray flux in the
intervening space, which therefore threads the secondary winding but not
the primary winding, stays constant, but the flux in the core depends on the
magnetic state of the core. This magnetic state is easily altered by tem
perature changes, mechanical shocks, changes in the source voltage for the
whole network and the passage of time. If one of the interposed shields is a
magnetic shield as well as an electric one, this situation is exacerbated. The
effect may well limit the accuracy to 1 in 10
7
.
The first three drawbacks can be overcome by appropriate arrangements of the
circuit, albeit with some complications. Figure 7.52 is coaxial apart from the two
triaxial leads from the calibrating transformer to the two adjacent taps of the
transformer to be calibrated. These leads should be routed closely together or
transformer to
be calibrated
1
0
1
2
3
4
5
6
7
8
9
10
D
calibrating
transformer
mag
ratio
Figure 7.51 A schematic diagram of a practical circuit for calibrating a voltage
ratio transformer
Transformers 231
twisted to keep their loop area small. The intermediate shields of these leads are
driven from taps on the magnetising winding of the calibrating transformer so that
their potentials are equal to those of the inner conductors. The capacitive loads on
these inner conductors are thereby reduced to an acceptably small value. The single
conductors between the shorted tap W and the outer conductors of the output ports
of the transformer being calibrated sense any small voltage between these outer
conductors. These small voltages need to be eliminated from the measurement
because the defined voltages of the transformer being calibrated are those found
between the inner and its individual outer conductor of each port. These small
voltages are transformed by the single conductors threaded through the toroidal
cores into the calibrating leads so that they are properly accounted for.
Thompson [3] overcame the fourth drawback by eliminating the calibrating
transformer and replacing it with a separate single winding section of the trans
former to be calibrated, which is identical to the other ratio sections but not con
nected to them. The capacitances between this section and the other sections are
accounted for in a particularly elegant way.
7.4.3 Calibration with a 1: 1 ratio transformer
Voltage ratio calibrations with uncertainties of the order of 1 in 10
9
are fairly readily
achievable by the straddling method about to be described, but realising the full
accuracy requires careful attention to the topology of the circuit and its electric screens.
1
0
5
10
1
10
generator
D
reference signal to port 1
J
4 J
1
J
3
J
2
supply
transformer
1
2
W
Wagner
10
1
0
5
3
4
4
calibrating transformer
mag
ratio
transformer
to be calibrated
Figure 7.52 A coaxial screened circuit for calibrating a voltage ratio transformer
232 Coaxial electrical circuits for interferencefree measurements
The principle of the method is to replace the somewhat uncertain primaryto
secondary ratio of the calibrating transformer by the stable 1:1 ratio of a bifilar
secondary winding (see section 7.2.3). The calibration network now has two loops
and two detectors, as shown in Figure 7.53, which shows only the secondary
windings of the transformers. T
A
is the transformer being calibrated; n m, n and
n m are three successive taps, which are nominally equally separated in voltage.
T
C
is the 1:1 calibration transformer. The complex quantities d
n
U
n
, d
C
U
C
and a
n
U
represent small inphase and quadrature voltages. The error d
C
U
C
in the 1:1 ratio
can readily be eliminated by reversing the 1:1 connections.
By altering its primary voltage, the total voltage 2U
C
d
C
U
C
developed by T
C
is adjusted to be exactly equal to the voltage U
n
(2 d
n
) between the n and n m
taps of T
A
. Then the detector D
2
indicates zero. At the same time, the voltage dif
ference between the middle taps of the two transformers is balanced and measured by
adjusting the small inphase and quadrature generator a
n
U to null the detector D
1
.
These balances are achieved simultaneously by a quickly converging iteration.
When observing D
2
, the comparatively large impedance of D
1
, which is arranged to
have a greater impedance than the whole outer loop, ensures that D
2
responds
primarily to the desired condition of equal total loop voltage. D
2
is shorted whilst
bringing D
1
to a null.
It is important to realise that the impedance of the loops to the source, a
n
U, is
only the small leakage inductance and series resistance of the windings and loops.
Hence, large currents will flow as a result of a small lack of balance between the
voltages; this effect gives the method great sensitivity.
First, assume that m1, i.e. the taps are adjacent.
When both detectors are nulled,
U
n
2 d
n
U
C
2 d
C
This is the auxiliary balance condition that ensures that neither transformer is
delivering current.
Also,
U
n
1 d
n
a
n
U
n
1 a
n
U
C
1 d
C
U
C
7:6
U
C
(1 + d
C
)
U
C
a
n
U
U
n
(1 + d
n
)
U
n
n
n m
n + m
T
C
T
A
D
1
D
2
Figure 7.53 A straddling method of calibrating a voltage ratio transformer
Transformers 233
for the main balance condition of no current in D
1
. In practice, a
n
U
n
can be
replaced by a
n
U to sufficient accuracy.
An equation similar to (7.6) is obtained by reversing the connections between
T
A
and T
C
at the defining output connectors of T
A
and rebalancing:
U
n
1 d
n
a
0
n
U
n
1 a
0
n
U
C
U
C
1 d
0
C
7:7
From (7.6) and (7.7), neglecting products of small quantities,
d
C
d
0
C
2a
0
n
a
n
and
d
n
a
n
a
0
n
d
C
d
0
C
2
If the calibration transformer ratio is independent of the connections to the
circuit, d
C
d
0
C
and then
d
n
a
n
sum 7:8
where
a
n
sum a
n
a
0
n
The recurrence relation ((7.8)) obtained by repeated observations a
n
as T
C
is
stepped along the taps of T
A
enables any ratio between the intertap voltages to be
found. In particular, the departure of the transformer from a nominal 10:1 ratio is
0:1
10
i0
10 ia
n
sum 7:9
If only the 10:1 ratio is needed, then one can employ a scheme involving four,
instead of ten, measurements where m=1, with a consequent reduction of error.
For example, one could use the three taps of T
C
to connect successively to the 1,
0, +1 taps, then the 1, +1, +3 taps, then +1, +3, +5 taps and finally to the 0, +5,
+10 taps, with the observed injected voltages a
1
(sum)U
C
, a
5
(sum)U
C
, a
10
(sum)U
C
,
respectively.
Figure 7.54 illustrates this scheme. The departure from a true 10:1 ratio is
0:16a
1
sum 4a
3
sum 2a
5
sum a
10
sum 7:10
234 Coaxial electrical circuits for interferencefree measurements
T
C
T
A
1
0
1
3
5
10
U
0
(1 + d
0
)
U
0
U
1
U
1
(1 + d
1
) U
3
U
3
(1 + d
3
)
U
5
U
5
(1 + d
5
)
U
10U(1 + d)
a
5
U
D
Figure 7.54 A scheme for calibrating a 10:1 voltage ratio
Many other measurement schemes are possible, and a good way to look for
systematic errors in the measurement is to measure the same total ratio by two
schemes.
We now examine the various precautions necessary to realise this method
accurately.
The property required of the calibration transformer T
C
is that the nominal
1:1 ratio of its secondary windings shall be constant and independent of the
overall potential of the ratio winding with respect to the rest of the circuit. Its
electric screening and the screening associated with the detector D
1
are critical in
meeting this requirement. The ratio must also be immune to the effect of any other
external perturbations, for example, vibrations or external electromagnetic fields,
especially at the calibration frequency. T
C
is not required to have an especially low
output impedance. A possible design is shown in Figure 7.55.
primary
inputs
calibration
outputs
A B C D E F G
Figure 7.55 A section through a 1:1 calibrating transformer
Transformers 235
B is a highpermeability toroidal core wound with the primary winding A. The
cross section of B and the number of turns of A are chosen to provide a sufficiently
high input voltage rating and input impedance. C is a second core placed adjacent
to A, and both are enclosed in a toroidal magnetic and electric primary shield D.
The secondary electric shield E surrounds D and the twisted bifilar 1:1 ratio
winding F is wound on it, using sufficient turns to generate the maximum cali
brating voltage needed. The ends are connected to form the three taps, which are
then led out in separate coaxial screens through the magnetic and electric screen G,
which surrounds the entire structure. Two adjacent tubular peripheral limbs (shown
on the right) from this shield enclose the ends of the primary winding separately.
By always maintaining the screens E and G at the potential of the centre tap of
the bifilar winding, the 1:1 ratio of the bifilar windings is unaffected by being
raised to various elevated potentials with respect to anything outside the space
enclosed by E and G. That is, raising the potential of E and G and the enclosed
bifilar winding with respect to the primary winding B and its shield D, or to
external circuitry, will not affect the 1:1 ratio.
7.4.4 The bridge circuit and details of the shielding
The bridge circuit and shielding are shown in Figure 7.56, configured for measur
ing the 0, 5, 10 ratio. Detectors (not shown) denoted by D
n
are considered to be
connected to the ports labelled n, although in practice, only a single detector
plugged successively into these ports is necessary.
By looking at this figure, it can be seen that the secondary shield of T
C
is
extended from T
C
to enclose totally the conductors from the taps of T
C
up to the
points A
1,
A
2
and A
3
where the output terminals of the transformer to be calibrated,
T
A
, are defined as those between inner and outer. A small overlap of this shield
with the outer conductors of the taps of T
A
ensures complete electric shielding.
This shield is elevated to the potential of the centre tap of T
C
by being connected to
the output of IVD I
4
, a very small quadrature adjustment being provided by IVD I
0
4
,
the phase change circuit and injection transformer J
4
. The potential is correct when
detector D
3
indicates zero, showing there is no current in the conductor connecting
the shield to the centre tap of T
C
.
As suggested in the diagram, the main detection transformer J
1
surrounds the
inner conductor from the centre tap of T
C
and its shield to A
2
. Since this shield is
(in general) at an elevated potential, capacitive currents will flow from it to the
shield of J
1
and those from the left of the split in the screen of J
1
will thread J
1
and
register on D
1
. By shielding the left portion with a second screen connected by a
wire, which passes around the outside of J
1
, the capacitive currents going from the
left and right sides to the right and left sides of the split screen of J
1
can be
equalised until D
1
registers a null. Lack of complete equality leads to a finite
detector indication. The extent of this effect can be measured, in order to make a
correction for it, by temporarily connecting at the middle tap of the three taps of A
3
a multiport connector in which the inners are brought together effectively at a
point, as are the outer conductors. See Figure 7.57 for a suitable construction in
236 Coaxial electrical circuits for interferencefree measurements
1
0 5
1
0
1
0
g
e
n
e
r
a
t
o
r
D
r
e
f
e
r
e
n
c
e
s
i
g
n
a
l
t
o
p
o
r
t
s
1
3
s
u
p
p
l
y
t
r
a
n
s
f
o
r
m
e
r
A
1
A
2
A
3
J
2
J
1
J
3
J
4
J
5
J
6
T
C
B
1
B
2
I
3
I
4
I
5
I
3 '
J
2 '
2
'
I
4 '
I
5 '
1
2
3
d
e
f
i
n
i
n
g
p
l
a
n
e
P
C
t
r
a
n
s
f
o
r
m
e
r
t
o
b
e
c
a
l
i
b
r
a
t
e
d
c
a
l
i
b
r
a
t
i
n
g
t
r
a
n
s
f
o
r
m
e
r
F
i
g
u
r
e
7
.
5
6
A
p
r
a
c
t
i
c
a
l
c
i
r
c
u
i
t
t
o
i
m
p
l
e
m
e
n
t
t
h
e
s
t
r
a
d
d
l
i
n
g
m
e
t
h
o
d
f
o
r
c
a
l
i
b
r
a
t
i
n
g
a
v
o
l
t
a
g
e
r
a
t
i
o
t
r
a
n
s
f
o
r
m
e
r
e
q
u
i
p
p
e
d
w
i
t
h
d
e
f
i
n
i
n
g
t
r
a
n
s
f
o
r
m
e
r
s
J
2
a
n
d
J
2
0
Transformers 237
which the inners are soldered symmetrically to a disc. This connector provides two
further ports at the potential of A
2
, and the other leads of the calibration transformer
are removed from A
1
and A
3
and plugged into these. Also, the supply to the pri
mary of T
C
is reduced to zero without altering the network configuration. Any
remaining small detector indication can be nulled with I
3
and I
0
3
, and the readings
thus obtained are corrections for this zero error to the actual readings obtained
when the bridge is reconnected as drawn.
Additionally, all three conductors to the right of J
1
are enclosed in a second all
embracing screen (not shown in Figure 7.56) also connected to P so that capacitive
currents, which would have flowed to the individual screens, are intercepted and
prevented from inducing error voltages in the inner conductors via the mutual
inductance between inner and outer conductors.
The conductors from T
C
are run closely together (to avoid emfs induced by
external magnetic fields) until they separate at point A
2
to go to the taps of T
A
at
A
1
, A
2
and A
3
. An additional wire between the outers of the A
1
and A
2
sockets runs
alongside the cable, and both wire and cable thread the highpermeability core E
1
once. There is the same arrangement between A
3
and A
2
. By this artifice, any
potential existing between the outer conductors at A
1
and A
3
is sensed and injected
into the inner conductors by the cores so that the circuit correctly responds to the
difference in potential between inner and outer of each of the various taps of T
A
.
The defining conditions of zero current at the output connectors of T
A
are
fulfilled at the gap in the outer screens even though the total current detection
transformers J
2
and J
0
2
are located a short distance away from the connectors
because they indicate zero total current in inner and screen. The capacitive current
between inner and screen to the left of the defining plane of the transformer up to
the gap in the outer must then have been supplied from the right along the inner
conductor and does not constitute a load on T
A
.
7.4.5 The balancing procedure
With reference to Figure 7.56, D
3
is first nulled by adjusting I
4
and I
0
4
. Then D
2
and
D
0
2
are nulled by adjusting I
5
and I
0
5
to alter the overall intertap voltage of T
C
to
match that of T
A
. Fine adjustment using continuously variable capacitors can help
if operation of I
5
and I
0
5
gives undesirable surges. Lack of a simultaneous null of D
2
disc
Figure 7.57 A multiport connector where inner conductors are brought together
at a point
238 Coaxial electrical circuits for interferencefree measurements
and D
0
2
is due to unequal capacitances of the cables from T
C
to the defining
transformers at A
1
and A
3
, and this can be eliminated by adjusting a small capa
citance trimmer C when measuring the 1, 0, 1 tap combination. C may be
required to be in the other outer straddling lead. D
2
and D
0
2
are then replaced by a
temporary short for rapid convergence, and D
1
is nulled by adjusting I
3
and I
0
3
.
Successive iterated balances of D
2
, D
0
2
and D
1
rapidly lead to simultaneous nulls
being registered.
As we indicated when outlining the principle of the method, it is necessary to
invert the 1:1 ratio and remeasure, to eliminate the small but finite departure of
T
C
from a perfect 1:1 ratio. This inversion is accomplished by interchanging
leads at A
1
and A
3
and reversing the current in the primary of T
C
by interchanging
the input and the shorting plug at B
1
and B
2
. The connection of the primarys screen
is unaltered by this reversal.
A check on the correctness of the method can be undertaken as follows. When
A
2
is connected to the zero potential tap of T
A
, it is also possible to connect two
nominally equal capacitances by means of Tconnectors at A
1
and A
3
to form
components of a 1:1 bridge. The actual ratio of the capacitors can be eliminated by
reversing the connections of this capacitance bridge at A
1
and A
3
and rebalancing.
The mean of these observations should equal the departure from nominal of the
ratio of the 1:1 calibrating transformer. Then if the same value of d
C
is obtained
from the transformer calibration network for other taps when A
2
is not at zero
potential, we can be confident that the calibration network is giving accurate
results.
The circuit of a suitable capacitance bridge is shown in Figure 7.58. The
relative values of the capacitors must be stable over the time of observation to the
accuracy required, which may be 1 in 10
9
or better. A pair of 1nF gasdielectric
capacitors, which have nearly equal temperature coefficients and which are in the
same temperaturecontrolled enclosure, would be suitable.
D
C
1
C
2
to bridge
source
A
3
A
1
Figure 7.58 A capacitance bridge to check the accuracy of the straddling method
Transformers 239
7.4.6 Calibrating voltage transformers by permuting
capacitors in a bridge
Another method of deriving ratios of voltage transformers has been described by
Cutkosky and Shields [4]. This method can measure a voltage ratio nominally equal
to the ratio of two integers m/n. Figure 7.59 illustrates the principle. If all the
admittances are nominally equal, the detector is near a null and can be nulled
exactly by the small emf U
1
injected in series with, say, U
n
. That is,
U
n
U
1
Y
1
Y
2
Y
n
U
m
Y
n1
Y
n2
Y
nm
7:11
U
m
D
U
1
U
n Y
1
Y
2
Y
3
Y
n+1
Y
n+2
Y
n+3
Y
n+m
Y
n
. . .
. . .
Figure 7.59 Principle of the permuting capacitors method for calibrating a
voltage ratio transformer
In Figure 7.59, by cyclic permutation, by transferring an admittance at each
step to the other side of the ratio arms as indicated by the arrows, a series of n m
similar relationships to (7.11) can be arrived at
U
n
U
2
Y
2
Y
3
Y
n1
U
m
Y
n2
Y
n3
Y
1
etc.
On adding the m n relationships we have, since each individual admittance
has voltage U
n
across it n times and voltage U
m
across it m times,
U
n
n
nm
i1
Y
i
nY
mean
nm
i1
U
i
U
m
m
nm
i1
Y
i
where we have assumed that SDY
j
U
j
is a negligibly small sum of products. DY
j
is
the difference Y
j
Y
mean
. Since
n mY
mean
nm
i1
Y
i
and n mU
mean
nm
i1
U
i
U
m
U
n
1
U
mean
U
n
_ _
n
m
7:12
The above analysis has assumed that U
n
and U
m
are not greatly perturbed by
the loading effects of the Y
i
and their associated connectingcable shunt
240 Coaxial electrical circuits for interferencefree measurements
admittances. The complex arithmetic needed to account for this loading can be
avoided by making the shunt admittances equal, for example, by Tconnected
shimming capacitors. There still remains the effect of the loading on the transfor
mer ratio, and this has to be accounted for either by extrapolation to zero load (see
section 5.3.9) or by using separate current and potential leads. A suitable circuit is
shown in Figure 7.60 for the usual case of determining a 10:1 ratio. Adjusting the
secondcore excitation and the adjustable source connected to the current winding
nulls the detectors D
1
and D
2
associated with the defining transformers. Then no
current is drawn from the potential winding. From the centres of these defining
transformers through the symmetrical fanout disc to the high port of the capacitors,
the shunt admittances must be made equal, as above.
The permuted admittances are usually capacitors as these components have least
change of value with applied voltage. Resistors are not so suitable because selfheating
alters their value with applied voltage more than capacitors, and they generate noise.
Selfinductors are not precise standards for the reasons given in section 6.4.13.
The individual capacitors do not need to be treated as fourterminalpair devices
with respect to the detector connections unless the differences between the values of
the capacitors, which this method also yields, are required because the detector cables
and connections remain unchanged throughout permutation of the capacitors.
The voltage coefficients of the capacitors are needed to correct for the effect of
changes in voltage they experience during permutation; these coefficients have to
be separately measured by one of the methods described in section 6.4.8.
The permuting capacitors technique is particularly appropriate for determining
the ratio of a voltage transformer at higher frequencies than a few kilohertz, and a
current source
aU jbU
D
to detector
to detector
10 capacitors
(only 4 are shown)
to 2nd core
IVD
0
2nd core excitation
to U
M
10U
M
U
M
U
s/b multiport
Figure 7.60 A practical circuit to implement the permuting capacitors method
Transformers 241
suitable shielded housing for the capacitors has been described by Awan et al.
[5] and Sedlacek et al. [6].
The arrangement described by Awan et al. [5] minimises the inductances in series
with the capacitors. A cross section through the device is schematically illustrated in
Figure 7.61. The 11 capacitors C
1
C
11
are NegativePositiveZero (NPO) ceramic
chip capacitors of nominally 50 pF value and are each two 100pF capacitors in series
selected from a large batch by combining small and large values in pairs so that the
series capacitances are closely equal. The small size (3 mm 1 mm 0.5 mm),
remarkable stability, low phase angle (7 10
5
rad), small temperature coefficient
(0 30 ppm/K) and negligible temperature hysteresis and voltage coefficient of these
capacitors make them especially suitable for this application. For the higher fre
quencies, 110 MHz, 5pF capacitors are more suitable to minimise the effect of
residual inductances in the device.
B S
10U
U
C
1
C
S1
P
A
C
11
C
S1
B
D
Figure 7.61 An arrangement of permuting capacitors suitable for frequencies up
to 10 MHz [5]
C
1
C
11
are connected to a central point (to which the detector D is connected)
in a circular configuration similar to the spokes on a wheel to ensure the geome
trical equivalence of each capacitor. The spokes end on two concentric rings of
SMB connectors. Ten of the capacitors can be connected in parallel to a circular
copper plate P by inserting shorting plugs S. The centre of P is connected to a
socket to which a bridge voltage of U can be applied. The remaining capacitor not
so connected (C
1
in the figure) has the voltage 10U applied to it, as shown. The
detector sides of C
1
to C
11
are shielded from P by a plate A, which has vertical fins
(not shown) to also shield the detector connections of each capacitor from one
another. The equality of stray capacitances C
S1
to C
S11
ensured by their symmetry
means that their only effect is to alter slightly and equally the effective values of C
1
to C
11
. The exposed outer ring of connectors and the shorting plugs are shielded
from the environment by caps B.
The small dimensions of the device and the proximity of go and return currents
within it ensure that inductive effects should not affect the ratio by more than a few
parts per million at frequencies of a few megahertz.
7.4.7 Calibration of current transformers
We do not discuss the calibration of these devices in general. A specific need will
be addressed in section 9.1.6. We note, however, that as a consequence of
242 Coaxial electrical circuits for interferencefree measurements
reciprocity, a calibration method that is suitable for a voltage ratio transformer can
also be applied to its dual device, a current ratio transformer, at least in principle,
by interchanging the roles of detectors and sources.
7.4.8 Assessing the effectiveness of current equalisers
This aspect of impedance comparison networks has caused more difficulty than
most others. It is important to realise that the equalisers in a network constitute a
component of that network whose correct functioning together is as important for
accurate results as any other component such as ratio transformers. Their function
and the method of measuring their effectiveness have already been described in
sections 3.2.1 and 3.2.2.
References
1. Cutkosky R.D. Active and passive direct reading ratio sets for the comparison
of audiofrequency admittances. J. Res. NBS 1964;68c:22736
2. Kibble B.P., Rayner G.H. Coaxial AC Bridges. Bristol: Adam Hilger Ltd.;
1984. (Presently available from NPL, Teddington, TW11 0LW, U.K.)
3. Thompson A.M. Precise calibration of ratio transformers. IEEE Trans.
Instrum. Meas. 1983;IM32:4750
4. Cutkosky R.D., Shields J.Q. The precision measurement of transformer
ratios. IRE Trans. Instrum. 1960;I9:24350
5. Awan S.A., Kibble B.P., Robinson I.A. Calibration of IVDs at frequencies
up to 1MHz by permuting capacitors. IEE Proc. Sci. Meas. Technol. 2000;
147(4):19395
6. Sedlacek R., Kucera J., Bohacek J. A new design of permuting capacitors
device for calibration of 10:1 highfrequency inductive voltage dividers.
CPEM 2006 Digest 2006;47677
Transformers 243
Chapter 8
General considerations about impedance
comparison networks
In this chapter, we attempt to give some guidance on the design and use of
impedancecomparing networks (bridges) in general. Specific examples of various
networks are described in chapter 9.
8.1 Designing bridge networks
Fulfilling the defining conditions of the impedances will often result in a network
almost designing itself, but sometimes some subtlety is needed to obtain conditions
that are provably equivalent to these defining conditions, as in the employment of
combining networks.
Elegance, simplicity and a return to first principles usually obtain a much better
result than resorting to ad hoc bolton solutions to remove the imperfections of the
original concept. Then, not only should there be greater satisfaction with the final
design and its operation but also attaining the desired accuracy will be more likely.
We find it better to separate the functions of isolation and generation of a
precise voltage by having an isolation transformer separate from the ratio trans
former, as shown in Figure 9.2 and subsequent examples. It also makes the prin
ciples of network design, in particular the Wagner circuit, clearer. See, for example,
the use of a separate current supply to ensure that no current is drawn from the two
potential ports of the ratio transformers in the networks described in sections 9.1.5,
9.2.3, 9.3.6, 9.3.7, 9.4.3 and 9.4.4. This design principle is a departure from the
designs by Kibble and Rayner [1], which are based on a combined isolation/ratio
transformer.
8.1.1 Applying coaxial techniques to classical
singleconductor bridges
There is a considerable and valuable number of older bridge circuits [2] that are
presented as singleconductor networks, often without much regard for interactions
between their different parts, which can readily be converted to a coaxial network
with its concomitant advantages. Singleconductor networks can conceptually be
converted by taking one reference point, usually termed earth or ground, and
topologically stretching and extruding it as tubes over all the single conductors,
components, detectors and sources. When current equalisers are added to each
separately identifiable mesh, the result is a coaxial network.
Converting the familiar fourarm bridge to coaxial form, however, leads to the
existence of large, but definite, shunt impedances from either end of every com
ponent to the coaxial outer conductor. The means of eliminating the effect of these
shunt impedances from the bridge balance condition provides an example of the
auxiliary balances needed in highaccuracy coaxial networks. These auxiliary
balances have been described in section 5.1.
Figure 5.1 shows a singleconductor fourarm bridge, which is redrawn in
Figure 5.2 as a completely screened network. Variable twoterminal impedances
that add to the shunt impedances already present have been added to the four
junction nodes between the components. To appreciate the purpose of these com
ponents, it is helpful to redraw the network of the outer conductors as a simplified
line, as shown in Figure 5.3.
8.1.2 Placement of current equalisers
It might be supposed that provided each and every mesh of a network has one and
only one current equaliser, there would be considerable freedom in choosing
exactly which cables to put them in. It is usually found that where there is a choice,
it is better to place equalisers in cables whose inner conductors carry the largest
currents.
But there are some further constraints for example, in impedancemeasuring
networks, placing a greater length of cable needed to make an equaliser in defining
leads should be avoided as far as possible because of the greater and therefore more
uncertain lead corrections needed (see section 5.4). Unfortunately, it will usually be
found that one equaliser must be placed in a defining cable, and the additional
length of cable and consequent cable correction accounted for.
Placement is assisted if the network diagram is drawn as single lines repre
senting the cables that join at the nodes, which are usually the bridge components.
If there is a Tconnection into a cable, it can conceptually be slid along the cable
until it coincides with a component. This does not alter the topology of the network.
The final placement can be checked by successively reducing the network by
one mesh, containing one equaliser, at a time, until nothing remains, and double
checked by recourse to the theorem stated in section 3.2.2.
8.1.3 Wagner circuit (and when it is applicable)
The bridge described in section 9.1.4 is a fourarm bridge, which has a ratio
transformer forming two of its arms. A required defining condition for reproducing
the ratio of the transformer accurately is that there should be no current flowing
through the short on the tap, as indicated in Figure 9.4. This condition is achieved
by adjusting R
W
and C
W
, each of which might need to be on either side of the
supply transformer. A property of a properly constructed ratio transformer is that its
voltage ratio is hardly affected by shunt impedance loads at the output ports at the
upper and lower bridge nodes, so the effect of these shunt impedances on the
246 Coaxial electrical circuits for interferencefree measurements
balance condition is often less than the required accuracy and no auxiliary detector
network balance is required.
As a matter of good design, it is better to first eliminate all low and variable
impedances to screen, if possible (e.g. inductive voltage dividers having a short to
screen in one port) as adjustable auxiliary balance components because the Wagner
balance would have to account for their lossy and variable input impedance. Ide
ally, there should be only one direct connection between the inner and the outer
conductors of the coaxial bridge circuitry. Usually, this is a short in the zero port of
a ratio transformer, and the Wagner circuit is adjusted so that there is no significant
current through it.
8.1.4 Convergence
In designing bridge networks, adjustable sources of current or of voltage are often
needed. To obtain satisfactory balance convergence, it is important to distinguish
between the uses of these dual sources. In general, quasicurrent sources of high
internal impedance are required when a current is to be injected in parallel into a
network node to augment or cancel an existing current. Quasivoltage sources of low
internal impedance are required when an adjustable voltage is to be injected in series.
A complete analysis of a bridge network to establish its convergence properties
analytically is usually very complicated and, when completed, difficult to interpret.
Fortunately, for the bridge networks described in this book, some general con
siderations are usually adequate. These networks usually consist of quasivoltage
sources such as IVDs, which are themselves connected across a voltage source, and
voltage injection transformers and quasicurrent sources where a comparatively
high impedance is connected to the output of a quasivoltage source with the
objective of injecting a current into a network node.
Because quasivoltage or quasicurrent sources have only low or high rather
than strictly zero or infinite output impedances, respectively, the various adjust
ments needed to make auxiliary and main balances in a network will never be
entirely independent. Sufficiently rapid convergence can usually be achieved
however, so that one, two or three iterations around these balances made in a
sensible order will be sufficient to achieve simultaneous balances to the accuracy
required. In general, it is better for current sources to have fixedvalue impedances
(capacitors or resistors) fed from variable voltage sources (IVDs).
The capacitances of a cable joining the C
L
ports of two fourterminalpair
impedances are part of a Wagner circuit, and the adjustment of a combining network
to eliminate the effect of the series impedance of this cable needs consideration. If the
adjustment is made by eliminating the effect of a voltage from an injection trans
former, all is well, but if the voltage is produced by inserting an impedance (a resistor
or an open circuit) at a point along the cable, the exact balance of the Wagner circuit
will depend on the position where the impedance is inserted. Therefore, to avoid an
awkward convergence problem, the value of the inserted impedance should be as
small as possible, consistent with obtaining the required sensitivity for the combining
network adjustment. A value of a few tens of ohms is typical.
General considerations about impedance comparison networks 247
Examples
1. A Wagner balance is intended to balance the currents from inner to outer
conductors on either side of a network through a designated zeropotential
point (a short between inner and outer conductors at some point in a network,
usually a tapping on a ratio transformer) (e.g. see Figure 9.2 and its accom
panying description a simple bridge with a Wagner arm). A current through
an adjustable high impedance (usually an adjustable capacitance and an
adjustable high resistance) between inner and outer conductors having high
potentials between them will best perform this adjustment. Because high
valued resistors have a large phase angle and are not therefore good AC
components, it is often better to use more moderate fixed values for compo
nents in series with the adjustable output of an IVD as in Figure 9.5. The
convergence between the main balance and the Wagner balance in most
bridges depends on the ratio of the output impedance of the potential ratio
windings to the impedance of the Wagner adjusting components. Therefore, it
is good practice to make the impedances that the Wagner components have to
balance as large as practicable by minimising the capacitances of cables hav
ing significant voltage between inner and outer (i.e. minimise their length) and
by avoiding having auxiliary IVDs with an output connected to screen.
The potential windings of the ratio transformer should be designed to have
minimal mutual output impedance by adopting one of the winding designs of
sections 7.2.27.2.5. In particular, because the impedance of the 0:1 winding
is reflected as the square of the turns ratio into the apparent output impedance
of the 0:10 winding by transformer action, the resistance of this section should
be about 1/100 of that of the 0:10 sections. Since the length of its winding is
1/10 of that of the 0:10 sections, its crosssectional area should therefore be
about ten times that of the conductors forming the 0:10 windings. The ordered
rope design in section 7.2.5 accomplishes this.
2. The voltages provided by the outputs of a ratio transformer need a small
adjustment to balance a bridge network. A quasivoltage source such as an
injection transformer in series with an output will best perform this task, as
described in section 7.3.3 and the accompanying text.
If, however, both the outputs of a voltage ratio transformer must be altered
without affecting their ratio, a current injected into an additional winding on a
second core will do this simply and accurately. This current should come from
a quasicurrent source, as described in Figure 7.60 and the accompanying text.
3. If a 10:1 voltage ratio transformer is constructed with an inbuilt isolated
primary winding or if it is calibrated with a specified isolating transformer and
interconnecting cables, no account need be taken of the defining condition
of zero current in the short in its zero tap. Satisfying the defining conditions of
zero current flow at the potential ratio taps is sufficient because the effect of
nonzero current in the zero tap short is part of the calibrated ratio. But if the
ratio transformer is calibrated independently of a separate isolating supply
transformer, as is suggested in this book, the condition of zero current in this
short should be satisfied by adjustable Wagner components.
248 Coaxial electrical circuits for interferencefree measurements
The reader should now be able to recognise other examples of convergence con
siderations in bridge networks described in this book and be able to successfully
design networks for other purposes.
8.1.5 Moving a detector to other ports in a bridge network
It is usual to employ only one detector in a coaxial bridge network. This detector is
connected in turn to the various ports where a detector needs to be nulled, as, for
example, shown in the various bridge diagrams in this book. The extra network
mesh introduced by connecting the detector should always contain the current
equaliser in the input detector lead so that no change in equalisation of the network
occurs. Usually, only the detector sensitivity need to be optimised (see section 2.1.1)
for the detector in the main balance position, as the other auxiliary balances are much
less critical.
8.1.6 Tconnecting shunt impedances for balance adjustment
In an ideal world, bridge network impedances would have extra ports connected
directly to their internal defining points for the purpose of adding small shunt
admittances to adjust their value slightly. Since existing impedance standards are
not in general provided with these extra ports, it is necessary to consider how to
connect a shunting admittance so that the shunt admittances and series impedances
of its cables have minimum effect and how, for measurements of the highest
accuracy, this effect can be calculated and accounted for. We assume, in this
section, that the shunting admittances to be added can be defined to sufficient
accuracy as twoterminalpair components (see section 5.3.6) and that since the
shunting admittances are usually small compared with the admittance of the shunted
component, cable corrections to the shunting admittance itself are negligible.
If a twoterminalpair component (corrected for the effect of its cables) is
shunted, the result is just the parallel combination of the shunted and shunting
impedances between the points where the Tconnections are made. This is an
example of the parallel connection of admittances described in section 5.7.
A fourterminalpair component is best shunted by a highimpedance
twoterminalpair component by placing the Tconnections in the C
H
and P
L
non
defining cables (Figure 8.1). The current drawn by the shunting connection in the
supply (C
H
) cable does not affect the definition of the shunted component, and
Z C
H
P
H
P
L
C
L
Z
Sh
Figure 8.1 Shunting a fourterminalpair impedance
General considerations about impedance comparison networks 249
although correction should be made for the small voltage drop down the P
L
lead
between the Tconnection and the internal defining point of the shunted compo
nent, this correction will often be negligible.
If the Tconnections are made into defining cables, it is necessary to consider
the effect of the lumped shunt load introduced by the shunting impedance and its
cables, and the situation is quite complicated.
8.1.7 Role of electronics in bridge design
In section 3.2.1, we have already described the use of amplifiers in feedback cir
cuits to augment the efficiency of current equalisers. Feedback amplifiers have also
been used to make some auxiliary balances automatically so that bridge balancing
is simplified. In these and other applications of electronics in bridge networks, it is
particularly important to consider the isolation of the electronic device and the
effect of any extra meshes it might introduce into the network.
8.1.8 Automating bridge networks
It is tempting to consider removing the tedium of repetitive rebalancing of a
bridge, by designing some degree of computercontrolled automation into the
various bridge balances. This is very difficult, partly because remotely operated
switches (relays) having the required degree of isolation between their magnetic
energising coil and their contacts are rare and tend to be very expensive, and partly
because writing a computer routine having the convergence properties conferred by
a human operator is not a simple task. Nevertheless, at least one successful system
has been achieved [3]. But, in general, it would seem that it is more efficient to
employ a manual bridge network designed to be as ergonomic as possible, perhaps
by employing analogue electronics to automate auxiliary balances (see section
5.1.1), and merely automate the recording and mathematical manipulation of the
results. The latter approach requires computerreadable outputs of the adjustable
components, and their isolation needs to be near perfect. An acceptable solution
might be to extend switch shafts via an insulating section into a separate screening
enclosure connected to the screen of the computer to operate within it switch wafers
appropriately connected to the computer.
8.1.9 Higherfrequency networks
Higherfrequency networks are necessarily smaller and can be simpler than their
lowfrequency counterparts, principally because measurement uncertainties are
greater and the consequent lower attainable accuracy is, nevertheless, sufficient for
present needs. Simplifications are as follows: (i) current equalisers will often be
unnecessary, (ii) extrapolation techniques can be used, which considerably simplify
the measurement networks and (iii) symmetry of the bridge networks can be
employed to substantially reduce offset errors.
250 Coaxial electrical circuits for interferencefree measurements
One key difficulty that emerges at higher frequencies, particularly at fre
quencies above approximately 0.5 MHz, is that the resistive and capacitive com
ponents used in auxiliary balance devices (such as current sources, combining
networks and Wagner balances) are no longer purely resistive or reactive but a
combination of resistance, capacitance and inductance. Therefore, it is useful to
measure the performance of these auxiliary components with a highfrequency
LCR meter first, to determine whether they are suitable, before incorporating them
in highfrequency bridge networks.
Sources in the megahertz frequency range normally have a 50W output
impedance. Sufficiently good impedance matching is needed when connecting such
sources to highfrequency networks, but networks having input impedances ranging
from few tens to few hundreds of ohms do not usually present a problem. High
frequency detectors, in particular lockin amplifiers, need extra care. Certain lock
in amplifiers operating near the edge of their frequency range can have reduced
performance in terms of gain, dynamic range and signaltonoise ratio. Unexpected
reference channel phase changes can occur because of the length of the connecting
cable, leading to longer setup times and increased bridge balance iterations
because the inphase and quadrature settings of the lockin amplifier do not match
those of the corresponding inphase and quadrature voltages in the bridge network.
However, this does not impact the accuracy of the bridge balance because attain
ment of zero detected voltage in both phases is independent of the phase of the
reference signal.
As in any branch of metrology, measurement reproducibility is the key for
debugging higherfrequency networks because then the effects of diagnostic
changes deliberately introduced into the network can be unambiguously observed.
8.1.10 Tests of the accuracy of bridges
Given good quality standards, it is relatively easy to assemble and balance a bridge
that, if proper attention has been paid to consideration of the signaltonoise ratio,
has a very great resolution. One can discern changes as small as 1 in 10
9
, with a
similar degree of stability of the balance. In this situation, it is rather easy to be
misled into believing that the results obtained also have a similar accuracy. There
are a number of general tests that should be applied to any untried network to
uncover possible causes of systematic error. For each test, the bridge is supposed to
be fully balanced in its normal state with the source emf and detector sensitivity at
the highest level at which the bridge network will be operated, unless otherwise
stated. We assume that the important injected voltages have been calibrated (see
section 7.3.6) and the equalisers are evaluated (see section 3.2.2).
A twoterminalpair component, which has zero admittance, is constructed as
illustrated in Figure 8.2, where a break in the inner conductor of a coaxial cable has
a complete screen across the break to intercept electric fields.
A twoterminalpair component, which has zero impedance, requires that no
voltage be generated across one port no matter how much current flows in or out of
General considerations about impedance comparison networks 251
the other. It is important not to include any mutual inductive coupling in the direct
impedance. Figure 8.2b illustrates a suitable arrangement.
Fourterminalpair versions of these twoterminalpair components merely
require the addition of extra ports by adding Tconnectors at either end.
Test Desired result
of test
Possible reasons why desired
result is not obtained
(1) Main bridge source
removed, replaced by
a short.
Main detector
reads zero
(Johnson noise
excepted).
Pickup of electric or magnetic fields
or a net current exists because of
inadequate isolation by transformer
screens or acoustic coupling.
(2) Detector gain decreased
from its highest setting.
No change of zero
detector reading.
Same causes as for (1).
Intermodulation distortion.
(3) Source output varied
by at least 2:1, keeping
the relative harmonic
content constant.
No change of zero
detector reading.
Voltagedependent transformer ratio
(demagnetise the transformer?) or
impedance having a voltage
dependence.
Intermodulation distortion.
(4) Source output constant,
harmonic content varied.
No change of zero
detector reading.
Frequencydependent bridge, inade
quate frequency selectivity of
detector.
Intermodulation distortion
(5) If a simple ratio
bridge, measure two
zero admittances.
Zero detector
reading.
Inadequate number or placement of
equalisers.
Auxiliary balances insufficiently well
understood.
(6) Measure two zero
impedances (source
matching may need
altering).
Zero detector
reading.
As for (5).
(7) A shorted turn around
the core of each
equaliser in turn.
A definite detector
deflection.
(i) Currents in inner and outer
conductors by chance balance in
that mesh retain the equaliser.
(ii) More than one equaliser in the
mesh. Remove all but one.
(8) Open coil (Figure 1.22)
from bridge source
passed near bridge
cables and components.
(A sensitive test; assess
what is a tolerably small
response. Very useful for
lowimpedance circuits.)
Zero detector
reading.
Inadequate magnetic screening of
components.
Meshes without equalisers.
(9) Metal plate (Figure 1.23)
connected to a high
potential from the
source, passed near
bridge cables and
components.
Zero detector
reading.
Cable outer conductors or component
cases are inadequate screens.
Unused coaxial ports without a
conducting cover.
252 Coaxial electrical circuits for interferencefree measurements
References
1. Kibble B.P., Rayner G.H. Coaxial AC Bridges. Bristol: Adam Hilger Ltd.;
1984. (Presently available from NPL, Teddington, TW11 0LW, U.K.)
2. Hague B. (revised by Foord T.R.) Alternating Current Bridge Methods.
6th edition. London: Pitman; 1971
3. Robinson I.A. et al. (Private communication)
I = 0
U
U = 0
I
(a)
(b)
Figure 8.2 (a) A twoterminalpair zero admittance. (b) A twoterminalpair zero
impedance
General considerations about impedance comparison networks 253
Chapter 9
Bridges to measure impedance ratios
We now bring together all the concepts and devices described so far to show how
they may be combined to make practical highaccuracy bridges. The number of
networks that may be devised is endless, and therefore, we have chosen just those
examples that either are of considerable practical importance to national standards
laboratories or illustrate the use of some device or concept. In principle, we show
how to relate like impedances of different values and then how to generate the
impedance standards of R, L and M from the unit of capacitance. The reader should
feel encouraged to devise other networks to serve other special needs.
It is certainly a major feature of the remarkable elegance of the subject that in
principle, bridge comparison networks can be built from just five dual pairs of
components. These we have introduced in an ad hoc manner, so that the sig
nificance of the pattern may well not have been apparent. The pairs are
1. impedances and admittances,
2. sources and detectors,
3. voltage and current ratio transformers,
4. injection and detection transformers and
5. combining networks and current sources.
We will often present a network as an admittance bridge; of course the distinction
between this and an impedance bridge is purely one of algebraic convenience.
Although it is possible to construct any network, and in particular any of the
classical fourarm bridges, in conductorpair form, we do not give examples of this;
instead, it seems more valuable to make use of the ready availability or ease of
construction of transformers of variable or fixed ratio to look at impedance mea
surement problems afresh and, starting from the defining conditions for the com
ponents to be compared, design appropriate networks, which make use of the
desirable properties of transformers.
Variable components, for example, decade resistance boxes, variable capaci
tors and switched tapped mutual inductors, were formerly used exclusively for
adjustable components. While these are still useful when fitted with coaxial term
inals, especially since many laboratories still have a good stock of them, we in
general advocate using a goodquality admittance of fixed value supplied from an
adjustable voltage source (an inductive voltage divider (IVD)) as the combination
has a greater stability against time and environmental effects.
In bridge networks, which need a detector in more than one position, only one
actual detector needs to be used. It can be connected into the various detector
positions in turn to make the associated adjustments.
9.1 Bridges to measure the ratio of like impedances
9.1.1 A twoterminal IVD bridge
Figure 9.1 illustrates a twoterminal bridge based on an IVD as a voltage ratio
device. This singleconductor network does not fall within the philosophy of this
book; it is discussed here only to establish the principle of the coaxial bridge, which
is described in the next section. A
1,2
(1 jD
1,2
) are the admittances to be compared.
When the detector is nulled, that is, the current through it is zero, the points either
side of it are at the same potential. With respect to the potential of the righthand
side of the detector, the potentials at the high terminals of A
1
and A
2
are then
(1 n)U and nU, respectively, being determined by the turns ratio of the IVD. We
can then equate currents at their low terminals.
U nUA
1
1 jD
1
jAk nU nUA
2
1 jD
2
A
A
2
n k
n
9:2
Equations (9.1) and (9.2) express an unknown A
1
in terms of known standards
A
2
and A.
For example, if components 1 and 2 are lossy capacitors having loss tangents
D
1
and D
2
, i.e.,
A
1;2
1 jD
1;2
jwC
1;2
1 jD
1;2
D
n
1 n
jA
A
1
(1 jD
1
)
A
2
(1 jD
2
)
k
1 k
U
Figure 9.1 The principle of a twoterminal bridge based on an IVD
256 Coaxial electrical circuits for interferencefree measurements
jA would then be a conductance G 1/R. If a small conductance is required, it can
be obtained from a Tnetwork (see section 6.4.10).
The bridge equations relate the capacitance values
C
1
C
2
n
1 n
9:3
and the loss tangents
D
1
D
2
wC
2
R
1
k n
n
9:4
The last equation represents a frequencydependent quadrature balance, but as
standard capacitors usually have very small loss tangents, this does not cause any
difficulty because this balance is much less critical than that relating the
capacitances.
Note the connection of the source and the IVDs to the high defining terminals
(see section 5.3.3) of the twoterminal components A
1
and A
2
. The current through
the components is drawn directly from the source, and the IVDs, being high input
impedance devices, draw little current. Hence, there is only a small potential drop
along the connecting wires between the IVDs and these terminals.
9.1.2 A twoterminalpair IVD bridge
This bridge, shown in Figure 9.2, is suitable for comparing the ratio of two like
admittances. It is a coaxial version of the bridge described in the previous section.
The direct connection at O between the detector and the IVD is replaced by the
indirect connection via the outer conductors of the coaxial cables.
D
1
2
jA
0
quad IVD
main IVD
C
W
R
W
supply
transformer
reference signal
generator
Figure 9.2 A twoterminalpair bridge based on IVDs
Bridges to measure impedance ratios 257
Notice that the shorted variable output of the lefthand IVD is now labelled as
the reference point 0. The small potential drops incurred by the currents returning
along the outer conductors to 0 from the point where the isolated detector is con
nected become part of the measurement and satisfy the twoterminalpair defini
tions of the components. Each mesh of the network has been provided with a
current equaliser to ensure the good definition of these currents and potentials.
Wagner components R
W
and C
W
balance the capacitances and associated
losses between the inner circuitry and outer conductors and shielding containers of
the components. R
W
and C
W
, either of which may need to be connected in either
position, are adjusted by replacing the short at port 0 by the detector and adjusting
them to obtain a null. The defining condition of the IVD, which demands that there
shall be no current through this short, is then fulfilled. This adjustment can be made
more conveniently by leaving the detector in its usual position and simply tem
porarily removing the short, adjusting R
W
and C
W
to again obtain a detector null,
replacing the short and iterating this procedure as necessary. (Once is usually
enough.)
The shunt admittances of the cables from the upper and lower output ports of
the main IVD to the components 1 and 2 are loads on the outputs of the IVD, which
affect the output voltages because of its finite internal winding resistances and
leakage inductances. To obtain greater accuracy, corrections must be applied for
these loadings. The corrections can be experimentally determined by connecting a
known shunt capacitance, which is an addition to the known cable capacitance, to
each IVD output in turn and extrapolating the observed change in the bridge bal
ance back to zero total capacitance. The Wagner balance will need readjustment to
account for each of the two added capacitances.
When accurate and calibrated IVDs are used in this simple bridge, it is capable
of a measurement uncertainty of less than 1 ppm for moderate values of Y
2
/Y
1
,
say, 10:1 > Y
2
/Y
1
 > 1:10.
9.1.3 A fourterminalpair IVD bridge
With but little extra elaboration, the network of the previous section can be adapted
to compare two like impedances defined in a fourterminalpair manner. This net
work is especially useful for comparing low impedances for which fourterminal
pair definition is necessary to obtain the required accuracy. By using the adaptor
described in section 5.3.10, two and fourterminal components can also be mea
sured on this bridge.
The circuit is shown in Figure 9.3. Often, in view of the other approximations
to be made, Wagner components will not be necessary. The theory of the Kelvin
double bridge given in section 5.6.1 is applicable, and we make use of the com
bining network described in section 5.6.2 to connect in series the two components
to be compared.
In this network, we also introduce a method of making the quadrature balance by
injecting in series with the main IVD zero output voltage a quadrature voltage deri
ved from a second IVD and phasechanging network as described in section 7.3.4.
258 Coaxial electrical circuits for interferencefree measurements
The bridge is balanced by first changing the switch position to make the value
of Z
k
small and adjusting the main and quadrature IVDs in turn to null the detector.
Then the switch is changed back to the position shown so that Z
k
is increased to be
of the same order of magnitude as Z
1
and Z
2
, and the detector is again nulled by
adjusting the settings of the combining network IVDs. This procedure is iterated
until a null is obtained whether Z
k
is inserted or not. The null obtained with Z
k
small
is the critical one, and the adequacy of this combining network auxiliary balance
can be tested with Z
k
small by changing the settings of the combining network by
an amount comparable with, but larger than, the fineness of their adjustment and
observing whether the main balance is significantly affected. The principle behind
this testing procedure is very important and should always be invoked to test the
satisfactory adjustment of the auxiliary balances of a bridge.
The chief approximations involved in this network arise from the voltage drops
along the cables to the main IVD and the current the main and quadrature IVDs
draw through the highpotential ports of Z
1
and Z
2
in violation of the defining
condition. The principal voltage drop arises from the magnetising current of the
singlestage main IVD flowing through the resistance and inductance of these
cables; this source of error (which causes large quadrature errors at low fre
quencies) can be largely overcome by using a twostaged IVD (see section 7.3.2)
with its magnetising winding driven directly from the source. There remains the
inevitable voltage drop caused by the current flowing through the shunt capacitance
of the cables and the series impedance R/2 jwL/2 of the cables (see section 3.2.6),
in addition to which the difference of the currents flowing through the shunt
admittances of these cables causes voltage drops in the output resistance and
leakage inductance of the IVD. The cables to the combining network have a very
small potential difference between inner and outer and do not contribute significant
error compared to the above causes.
D
R
W
C
W
Z
1
Z
2
Z
k
main IVD
Kelvin
quad IVD
supply
transformer
reference signal
generator
Figure 9.3 A fourterminalpair bridge based on IVDs
Bridges to measure impedance ratios 259
These errors limit the accuracy of the network to perhaps about 1 in 10
7
, but
the network is nevertheless useful for comparing low impedances. The resolution
obtainable is adequate as a detection transformer (section 2.1.4) can also provide
impedance matching between a low bridge impedance and the comparatively high
input impedance of a typical detector. The quadrature balance is provided by
injecting a quadrature voltage jbU in series with the IVD output, and the approxi
mate bridge balance equations follow straightforwardly from equating the division
of potential ratios across the components and across the IVD plus the injected voltage
if D
1,2
(1. They are, in the notation of section 9.1,
A
1
A
2
n
1 n
9:5
and
D
2
D
1
b
n1 n
9:6
9.1.4 A twoterminalpair bridge based on a 10: 1 voltage ratio
transformer
The comparison of two like impedances whose values are nominally in a 10:1 ratio
is a frequently encountered problem. A 10:1 voltage ratio transformer (see
sections 7.3.7 and 7.3.8), which has been calibrated (see sections 7.4.17.4.6),
is the ratio component in the bridge drawn in Figure 9.4, and the small additional
D
main injection transformer ratio transformer
R
W
C
W
Z
1
Z
2
supply
transformer
reference signal
generator
Figure 9.4 A twoterminalpair bridge based on a 10:1 voltage ratio
transformer
260 Coaxial electrical circuits for interferencefree measurements
variable voltage needed to balance the bridge can be obtained from an injection
transformer (see section 7.3.3 and 7.3.4). The whole bridge network is driven
by an isolating transformer (see section 1.1.5) to preserve the isolation of the
bridge.
The nominal 10:1 transformer ratio is defined as the ratio U
1
/U
2
where U
1
and U
2
are the voltages appearing across the upper and lower exit ports of the
transformer under conditions of zero current flow through these ports. For moderate
impedances Z
1
and Z
2
and cable shunt admittances, this ratio will not be greatly
affected by the finite current drawn from the coupled low output impedances of the
transformer. If the inaccuracy this introduces is unacceptable, a calculated correc
tion can be made for the loading effects, or the fourterminalpair values of Z
1
and
Z
2
can be obtained by the extrapolation technique (see section 5.3.9) at the expense
of performing extra balances.
The voltage presented to Z
2
is U
2
(1 a jb), where aU
2
and jbU
2
are the in
phase and quadrature injected voltages, respectively, so that the equation
U
2
1 a jb
Z
2
U
1
Z
1
9:7
holds when the detector is nulled.
Notice that the incorporation of a wellconstructed sourceisolating transfor
mer (see section 1.1.5) makes the connection of a phasesensitive detector easier
(see section 1.1.2).
9.1.5 A fourterminalpair bridge based on a twostage
10: 1 voltage ratio transformer
We now describe a network in which both the defining conditions for the trans
former output voltages and the fourterminalpair definition of the components to
be compared are satisfied. The complexity of the network is greater than that
described in the last section, but nevertheless, achieving the detector balances at the
various ports is straightforward.
The bridge is shown in Figure 9.5. The like impedances Z
1
and Z
2
are to be
compared: their nominal values are such that Z
1
& 10Z
2
. They are defined by the
fourterminalpair convention of section 5.3.8.
The allimportant ratio of the voltages across the impedances is measured
under conditions of zero current at the defining points of the impedances. The
voltage ratio is varied to balance the bridge by injecting a small, known voltage DU
from the injection IVD and associated injection transformer. Simultaneous zero
current at the output of the two potential windings as sensed by the detection
transformer is achieved by iterative adjustment of current sources 1 and 2.
At the lowpotential end of each impedance, the defining current leads are
joined with provision for injecting a voltage with an injection transformer con
nected to the generator and adjusting the Kelvin network to null the detector. After
making this auxiliary balance, the injected voltage is set to zero by removing the
Bridges to measure impedance ratios 261
Z
1
Z
2
s
u
p
p
l
y
t
r
a
n
s
f
o
r
m
e
r
W
a
g
n
e
r
K
e
l
v
i
n
i
n
j
e
c
t
i
o
n
I
V
D
t
o
d
e
t
e
c
t
o
r
t
o
d
e
t
e
c
t
o
r
t
o
d
e
t
e
c
t
o
r
t
o
g
e
n
e
r
a
t
o
r
g
e
n
e
r
a
t
o
r
D
r
e
f
e
r
e
n
c
e
s
i
g
n
a
l
d
e
t
e
c
t
o
r
p
o
r
t
r
a
t
i
o
t
r
a
n
s
f
o
r
m
e
r
c
u
r
r
e
n
t
s
o
u
r
c
e
1
c
u
r
r
e
n
t
s
o
u
r
c
e
2
F
i
g
u
r
e
9
.
5
A
f
o
u
r

t
e
r
m
i
n
a
l

p
a
i
r
b
r
i
d
g
e
b
a
s
e
d
o
n
a
t
w
o

s
t
a
g
e
1
0
:
1
v
o
l
t
a
g
e
r
a
t
i
o
t
r
a
n
s
f
o
r
m
e
r
262 Coaxial electrical circuits for interferencefree measurements
connection to the generator and replacing it with a shorting plug. Then, if neces
sary, DU is readjusted. Convergence is very rapid, and only a few iterations of these
balances are necessary to achieve a simultaneous null.
The detection transformers and associated detectors when nulled ensure that
the condition of zero current at the outputs of the potential windings is satisfied.
The injection IVD, phasechange circuit and injection transformer provide adjus
table inphase and quadrature voltages, respectively. By twostaging these (see
section 7.3.4), the necessary injection accuracy of better than 10
5
for compara
tively large values of DU/U is achievable.
The adjustable voltage sources are all driven from the firststage winding of
the ratio transformer to avoid loading the potential ratio winding of this
transformer.
9.1.6 Equalpower resistance bridge
In this elegant network, devised by Cutkosky [1], the same power is dissipated in
each of the resistors being compared. This avoids corrections for the load coeffi
cients of the resistors and optimises the signaltonoise ratio when the most sig
nificant noise contribution is their Johnson noise at room temperature. It is a
combination of a voltage ratio bridge in which the highervalued resistor con
tributes most noise with a current ratio bridge in which the noise of the lower
valued resistor dominates. It is fully described by Kibble and Rayner [2].
9.2 Bridges to measure the ratio of unlike impedances
9.2.1 RC: the quadrature bridge
The purpose of a quadrature bridge is to relate resistance R and capacitance C by
means of the basic circuit of Figure 9.6.
U
jU
1=jwC
R
1
jwCR
9:8
C
jU
U
R
D
Figure 9.6 The principle of a quadrature bridge to relate resistance and
capacitance
Bridges to measure impedance ratios 263
It is not practically possible to generate jU from U with anything like the
required accuracy, and the remainder of this section is concerned with the inge
nious solution to overcome this.
A second bridge, whose balance condition is
jU
U
1=jwC
2
R
2
1
jwC
2
R
2
9:9
is combined with the first bridge, as shown in Figure 9.7.
U can be generated from U with a 1:1 ratio transformer, and if the depar
ture from exactness of the transformer is eliminated by interchange of its outputs,
1 in 10
9
accuracy for this ratio is achievable.
U
R
1
R
2
C
2
C
1
D
1
D
2
jU
U
Figure 9.7 Combining two quadrature bridges
Now a way must be found so that a condition equivalent to a simultaneous null
of D
1
and D
2
can be obtained that does not depend on the exact value of jU. This
can be done if the potentials at A and B are combined via auxiliary impedances R
a
,
R
b
, C
a
and C
b
to be sensed by a single detector D. Applying linear network theory,
the response of D to jU by itself can be found if the sources U and U are set to
zero. When this is done, the whole network can be redrawn as a twinT bridge
(Figure 2.2). Remember that the short thick lines to which 0, R
a
, C
b
and jU are
connected represent network points that are connected together (by the outer con
ductors of coaxial cables in the eventual design). If all the capacitances and all the
resistances are nearly equal, the twinT balance condition wRC 1 can be achieved
by adjusting R
a
and C
b
. The network of R
a
, R
b
, C
a
and C
b
is called the detector
combining network.
Finally, jU can be generated from voltage sources 2U and 2U and the
adjustable components C
J
and R
J
, which again act in conjunction with the rest of
the network, as shown in Figure 9.9.
264 Coaxial electrical circuits for interferencefree measurements
D 0
C
J
R
J
R
3
R
a
R
1
R
B
C
b
C
2
C
4
C
A
jU
+2U
2U
+U
U
Figure 9.9 The principle of a complete quadrature bridge
The main balance condition
w
2
C
1
C
2
R
1
R
2
1 9:10
can be achieved approximately by altering the value of one of these components to
null the detector. Then the detector combining network can be radically unbalanced
by temporarily altering the value of one of its components and C
J
and R
J
adjusted to
again null the detector. This is termed the jsource balance. Then C
J
or R
J
can be
temporarily altered, so that their value can be reset after the detector combining
network is rebalanced to null the detector. This procedure, which has achieved the
approximate adjustment of the main balance and the product of two auxiliary bal
ances, should be iterated until all three balances are simultaneously achieved,
remembering that the accuracy depends only on the product of the two auxiliary
U
jU
D
R
1
R
2
R
a
R
b
C
b
C
a
C
1
C
2
U
Figure 9.8 The principle of a quadrature bridge coupled to a detector combining
network
Bridges to measure impedance ratios 265
balances. These, therefore, need each only be made to an accuracy of, say, 1 in 10
4
if 1 in 10
8
accuracy is to be achieved for the main balance.
For simplicity in the above explanation, we have assumed that C
1
, C
2
, R
1
and
R
2
are ideal, that is, they have no small quadrature component. In practice, they will
each have a small phase angle j and can be expressed as C
1
(1 jj
C1
), etc. On
inserting these corrected values into (9.8) and (9.9) and equating separately the real
terms, (9.10) is again arrived at if second and higher products of the j terms are
neglected. The imaginary terms yield the condition
j
C1
j
C2
j
R1
j
R2
0 9:11
where the small phase defect of the U/U transformer ratio has been eliminated
by taking the mean of forward and reverse transformer connections.
This network was devised principally for deriving the unit of resistance, the
ohm, from that of capacitance, the farad, as realised by a calculable capacitor. More
recently, this derivation has also been operated in reverse to derive the farad from
the ohm, as realised from the quantum Hall effect [3].
A quadrature bridge can also set up a standard of phase angle, starting from
two capacitors, known to have very low loss factors to derive the phase angle of
resistors, which is much more difficult to do otherwise. The principle is to take
three similar resistors, each of which is provided with an adjustable trim to alter
their phase angle as described in section 6.4.11 and connect them, two at a time,
into a quadrature bridge. The phase angle imbalance of the bridge reflects the sum
of the phase angles of the two resistors and the two capacitors (9.11). Let the known
or estimated sum of the phase angles of the two capacitors be f
C
, the phase angles
of the resistors be f
1
, f
2
and f
3
and the magnitude of the quadrature adjustment
required to balance the bridge be df. Then
f
C
df
a
f
2
f
3
0
f
C
df
b
f
3
f
4
0
f
C
df
c
f
4
f
2
0
Adding the first two equations and subtracting the third,
f
3
f
C
df
a
df
b
df
c
2
9:12
The quadrature adjustment of the bridge can be altered by f
3
, and the phase
angle trim on resistor 3 altered to balance the bridge. This procedure can be repe
ated using similar equations corresponding to the other two resistors, and the final
adjustment of all three checked, with two at a time in the bridge.
266 Coaxial electrical circuits for interferencefree measurements
9.2.2 The quadrature bridge a twoterminalpair design
This simple and elegant design was originally conceived by Thompson. The principal
components C
1
, C
2
, R
1
and R
2
were originally regarded as being twoterminalpair
components defined at the ends of their cables, which are therefore a part of
these components. The accuracy of the bridge is improved by treating the resistors as
fourterminalpair components, as shown in Figure 9.10. The limitation of the defi
nition of the resistors to a few parts in 10
8
by the lack of reproducibility of the contact
resistances of their connectors is removed. The capacitors cause no problems; the
lead and contact resistances of their terminating connectors only affect slightly their
phase angle. The cables of the twoterminalpair capacitors C
1
and C
2
stay with them
when they are transferred to other bridges, but account has to be taken of the added
effect on their values in this network of the cables internal to the resistors up to the
internal defining points of the resistors. The loading effect on the 1:1 ratio trans
former of the auxiliary 100pF capacitors and their cables on the highpotential ports
of C
2
and R
1
, if these cables are similar, is symmetrical and does not affect the result.
Use of these capacitors to determine the 1:1 voltage ratio applied to R
1
and C
2
avoids considering the change in loading effects on the ratio transformer if it were to
be reversed instead.
The potential jU at the junction of R
2
and C
1
is set up by the auxiliary adjus
table capacitor C
J
connected to the 2U tap and the adjustable resistor R
J
supply
transformer
generator
D
r
e
f
e
r
e
n
c
e
detector port
filter
to detector to detector
C
J
R
J
C
1
R
1
R
2
C
2
R
T1
R
T2
C
T1
C
T2
+2U
+jU
2U
0
+U
U
injection IVD
100 pF
100 pF
x
y
Figure 9.10 A twoterminalpair quadrature bridge
Bridges to measure impedance ratios 267
connected to the 2U tap. The detector combining network is C
T1
, C
T2
, R
T1
and
R
T2
. a
1
is the mean of the small inphase injected voltages needed to balance the
auxiliary 100pF capacitance bridge and with the capacitors together with their
connecting leads interchanged, and a
2
is the injected inphase voltage needed to
balance the quadrature bridge.
The balance procedure is to first break the connections at X and Y and connect
the bridge side of them together into the shorted centre tap of the supply/ratio
transformer, thus setting U to zero, and altering the components from the 2U
taps to null the detector to properly adjust the nominal jU voltage. Then, after
restoring the connections at X and Y, the jU adjustment is temporarily changed in
a restorable manner by, for example, shorting the variable capacitor C
J
, and the
detector network is balanced by adjusting C
T2
and R
T1
. Finally, the short across the
variable capacitor is removed, and the injected voltage is adjusted to again balance
the detector. The whole balance procedure is iterated enough times so that no
readjustment of a
2
is required to within the target accuracy. Whether the jU and
detector combining network balances have been made with sufficient accuracy can
be tested by altering in turn, for example, C
J
and C
T2
by a fraction, say, 1% of their
value. Since the bridge balance is affected only by the product of the setting
accuracy of these components, the bridge balance should not be affected by more
than 1% of the setting accuracy of the other.
The relationship
1 a
1
a
2
w
2
C
1
C
3
R
2
R
4
9:13
holds when the bridge is balanced.
9.2.3 The quadrature bridge a fourterminalpair design
Improved definition of the capacitances C
1
and C
2
can also be obtained by adopting
a fourterminalpair configuration for them at the expense of increased complexity
in the bridge design. A circuit has been devised and published by Cutkosky [1], and
the version that the authors have used is drawn in Figure 9.11.
The fourterminalpair components C
1
, C
2
, R
1
and R
2
are all joined in series by
combining networks Kelvin 1, 2 and 3 (see section 5.6.2) in a way that is equivalent
to realising their defining conditions. Networks Kelvin 1, 2 and 3 are adjusted so
that voltages DU
1
, DU
2
and DU
3
injected by temporarily connecting the generator
in turn to their corresponding injection transformers have no effect on the main
detector, thus producing in turn a condition equivalent to DU
1
, DU
2
and DU
3
being
zero. As in the previous section, C
T1
, C
T2
, R
T1
and R
T2
form a detector combining
network, which, together with the rest of the bridge, is a twinT bridge, which can
be adjusted by altering the values of the detector combining network so that the
main detector is immune to departures of the auxiliary source created by the
adjustable current source networks, R
J
and C
J
(see section 5.6.2) from the required
value jU. Only one source and one detector are required. They are plugged in turn
to the ports concerned to make the main and the auxiliary balances, and all ports
except the ones they are currently moved to are closed with shortcircuiting plugs.
268 Coaxial electrical circuits for interferencefree measurements
s
u
p
p
l
y
t
r
a
n
s
f
o
r
m
e
r
g
e
n
e
r
a
t
o
r
d
e
t
e
c
t
o
r
p
o
r
t
t
o
d
e
t
e
c
t
o
r
t
o
d
e
t
e
c
t
o
r
t
o
d
e
t
e
c
t
o
r
C
J
R
J
C
1
C
Q
R
1
R
2
C
2
C
P
R
T
1
R
T
2
C
T
1
C
T
2
i
n
j
e
c
t
i
o
n
I
V
D
W
a
g
n
e
r
D
r e f e r e n c e
r
a
t
i
o
t
r
a
n
s
f
o
r
m
e
r
c
u
r
r
e
n
t
s
o
u
r
c
e
1
c
u
r
r
e
n
t
s
o
u
r
c
e
2
f
i
l
t
e
r t
o
g
e
n
e
r
a
t
o
r
t
o
g
e
n
e
r
a
t
o
r
t
o
g
e
n
e
r
a
t
o
r
K e l v i n 1 K e l v i n 2
K
e
l
v
i
n
3
F
i
g
u
r
e
9
.
1
1
A
f
o
u
r

t
e
r
m
i
n
a
l

p
a
i
r
q
u
a
d
r
a
t
u
r
e
b
r
i
d
g
e
Bridges to measure impedance ratios 269
The defined potentials U are applied to the upper potential ports of C
2
and R
1
via defining transformers by the 1:1 ratio transformer, the condition I 0 being
achieved by adjusting the current sources 1 and 2 to null the detector when it is
connected to the corresponding defining transformer.
The detector combining network is adjusted by altering C
T2
and R
T1
to null the
main detector when either C
J
or R
J
is temporarily maladjusted. C
J
and R
J
are
adjusted by nulling the main detector when either C
T2
or R
T1
is temporarily
maladjusted.
The main balance adjustment is then made with the injection IVD, which, in
conjunction with C
P
and C
Q
, have the effect of putting small variable capacitances
in parallel with C
2
and R
1
.
The departure of the ratio transformer from a perfect 1:1 ratio is eliminated
by reversing the connections at the output ports of its associated detection trans
formers and at the inputs to it from the supply transformer. This also reverses the
sense of the injection IVD, so that their readings corresponding to zero voltage
applied to C
P
and C
Q
need not be determined. The defining condition of zero volts
at the centre port of the ratio transformer is obtained by adjusting the Wagner
components, so that the main detector is again nulled when the short there is
temporarily removed.
All these adjustments are iterated as necessary, but convergence is very rapid.
The main detector is coupled into the bridge via a harmonic filter (see section
2.1.2).
There are 20 meshes in the network shown, and each is provided with an
equaliser whose effect on the total network is evaluated by the method described in
section 3.2.2.
The voltage applied to C
P
is (0.5 k
1
)/2U (1 2k
1
)/U, where k
1
is the
setting of the associated IVD output, and hence, the admittance of C
2
is augmented
by C
P
(1 2k
1
), so that the bridge balance condition, (9.10), becomes
w
2
C
1
C
P
1 2k
1
R
1
C
2
R
2
1 9:14
On reversing the ratio transformer,
w
2
C
1
C
P
1 2k
0
1
R
1
C
2
R
2
1 9:15
and taking the mean of these two results, which eliminates any small departure
from a nominal 1:1 ratio of T
2
, we finally have for the inphase balance
condition
w
2
C
1
C
P
k
0
1
k
1
R
1
C
2
R
2
1 9:16
For the quadrature adjustment, the phase defect represented by the capacitance
C
Q
shunting R
1
is similarly modified by 2k
2
k
0
2
C
Q
, so that the relation of (9.11)
between the phase defects becomes
270 Coaxial electrical circuits for interferencefree measurements
j
C1
j
C2
j
R1
j
R2
wR
1
C
Q
k
2
k
0
1
0 9:17
This bridge has been constructed with components of nominal value C
1
C
2
1 nF, R
1
R
2
10
5
W and C
P
C
Q
1 pF. The nominal frequency was
w 10
4
rad/s.
With U 10 V, giving 1 mW power dissipation in the resistors and a detector
time constant of 3 s, a resolution and accuracy approaching 1 in 10
9
can be
achieved.
9.2.4 Bridges for measuring inductance
Most national standards laboratories calibrate inductors in terms of the traceable SI
values of capacitors by means of the fourarm MaxwellWien bridge illustrated in
Figure 9.12. The balance equations are
L
x
C
3
R
1
R
2
and
R
x
R
1
R
2
R
3
By choosing appropriate values of C between 10 pF and 10 nF and R
1
and R
2
between 10 W and 1 MW, Hanke et al. [4] calibrated their laboratory standard
inductors, which had values between 100 mH and 100 H at frequencies between
50 Hz and 1 MHz.
The basic network requires the addition of a shield and a Wagner network (see
section 8.1.3) to adjust the shield potential to that of the detector. Despite its sim
plicity, this bridge can attain a relative accuracy of about 10 ppm at lower fre
quencies, reducing to about 1% at higher frequencies. These accuracies are entirely
adequate for calibrating the standards of commercial laboratories and, in any case,
D
R
1
R
3
R
2
C
3
R
X
L
X
Figure 9.12 Principle of the MaxwellWien bridge
Bridges to measure impedance ratios 271
are limited to the extent to which inductance standards can meet their defining
conditions.
Cote [5] has constructed a coaxial fourterminalpair version of the Maxwell
Wien bridge, which attains only a similar accuracy despite greater complexity. He
also reports that the convergence of his network is poor, making the bridge some
what tedious to balance.
Sedlacek and Bohacek [6] have described a network, which calibrates induc
tance in terms of resistance, which is very closely related to the quadrature bridge
described in section 9.2.1 for relating capacitance to resistance.
Kibble and Rayner [2] describe a very simple resonance technique to measure
a highQ special inductor in terms of capacitance. Other values of inductance are
measured in terms of this inductor by an inductance ratio bridge.
9.3 AC measurement of quantum Hall resistance
By using the multipleseries connection scheme of a quantum Hall resistance
device (see section 6.7.1), AC quantum Hall resistance measurements can be made
as for any other twoterminalpair impedance standard. Also, because of the special
properties of the quantum Hall effect (see section 6.6.2), it can provide self
contained combining networks (see section 6.7.1). This simplifies the measuring
bridges, which compare quantum Hall resistance devices with each other or with
conventional impedance (resistance or capacitance) standards. These modified
bridge networks as well as some special bridge networks for the measurement of
the contact and longitudinal resistances to characterise AC quantum Hall resistance
devices are described in the following sections.
9.3.1 AC contact resistance
Contact to the twodimensional electron gas can be achieved by bonding wires to
lithographic AuGeNi contacts or by soldering thin wires to tin ball contacts. Non
ideal or poor contacts can cause two effects. First, the small residual currents
flowing through the potential contacts (as a consequence of the multipleseries
connection scheme) produce a small and calculable voltage drop at the contact
resistances, which must be taken into account. Second, poor contacts can cause a
systematic deviation of the Hall resistance from the quantised value, and this is
often accompanied by excess noise and irregular structures in the plateau. It is
therefore necessary to measure the contact resistances.
The AC contact resistance can be measured, for example, by a threeterminal
pair measurement, which is simple and fast and yields a resolution of about 0.01 W.
The scheme is shown in Figure 9.13. It yields the sum of the contact impedance and
the series impedance of the particular lead. This is just the connecting impedance,
which is relevant to the multipleseries connection scheme (see section 6.7.1). The
connecting impedance is mainly real and within a good approximation equal to the
sum of the contact resistance and the series resistance of the particular lead. As
measuring the series resistance of the individual quantum Hall resistance cables is
272 Coaxial electrical circuits for interferencefree measurements
part of the measurement of the cable parameters (see section 6.7.2), this can be
subtracted to find the actual contact resistance. See [30,31] of chapter 6.
If desired, it is also possible to determine the small imaginary part of the
contact impedance. This requires a phase adjustment of the detector and can
be achieved, for example, if a resistance of a few ohms, chosen to have a negligible
phase angle, is temporarily inserted into the shortcircuited lead. The phase of the
detector is adjusted so that the change of detector signal is in phase. The small
quadrature component of the connecting impedance can then be measured as the
quadrature reading of the detector. This quadrature component is usually mainly
caused by the series inductance of the particular lead.
The impedance of all contacts can be measured successively by reconnecting
the source, the detector and the shortcircuit plug to the appropriate terminals of the
quantum Hall resistance device. The applied voltage and the gain and sensitivity of
the detector can be measured by interchanging the source and detector. For litho
graphic AuGeNi contacts, the current applied to the potential contacts should be
considerably smaller than the current applied to the current contacts because the
potential contacts have smaller dimensions and tolerate less current and because
this is closer to the actual situation of the multipleseries connection scheme.
A contact is called ohmic if the contact resistance shows no current depen
dence. Good ohmic contact resistances are usually of the order of 1 W or less and
show no frequency dependence, so that the AC and DC contact resistances are
believed to be practically the same (on the evidence from only a few individual
typical AC quantum Hall resistance devices).
The AC contact resistances can also be measured as a function of frequency. If
resonances occur, this may indicate mechanical vibrations of excessive long bond
wires (see section 6.7.2).
9.3.2 AC longitudinal resistance
The AC longitudinal resistance is a key parameter for investigating the quality of a
quantum Hall resistance device and for understanding its AC characteristics. For
ungated and splitgated devices, all the frequency and currentdependent structures
in the plateaux of R
xx
apply in a very similar way to R
H
. The measurement of R
xx
cryostat
B
D
reference
~
U
Det
= UZ
k
/R
H
U = 0
U
Figure 9.13 Threeterminalpair measurement of the connecting impedance of an
individual quantum Hall resistance lead
Bridges to measure impedance ratios 273
yields information on the relevant capacitances as a function of the magnetic field
as well as on the frequency and current dependence of the associated AC losses.
Also, the homogeneity of the capacitive effects along the device edges can be
tested, and the effect of gates and shields can be investigated. Furthermore, the
individual measurements of the Hall and longitudinal voltages along a mesh with
three (or more) nodes can be tested for consistency. Invoking Kirchhoffs law, they
should add up to zero. Finally, the frequency dependence of the longitudinal
resistance can be extrapolated to zero as a test of whether or not the quantum Hall
resistance device meets the DC guidelines, without the need for a separate DC R
xx
measurement. Altogether, the AC measurement of R
xx
yields a lot of valuable
information. The measuring circuits for the low and highpotential side of a
quantum Hall resistance device are different and are therefore described in separate
sections 9.3.3 and 9.3.4, respectively.
9.3.3 Measuring R
xxLo
Figure 9.14 shows the equivalent circuit of a R
xx
measurement at the lowpotential
side of a quantum Hall resistance device. The highpotential side of the equivalent
circuit is the same as shown in Figure 6.42 and thus not explicitly shown in Figure 9.14.
The terminals of the quantum Hall resistance device selected for the R
xxLo
measure
ment are labelled A and B.
The admittance of the lead going to A in the presence of a nonzero potential at
the Aterminal drives an unwanted AC current through the Aterminal, and the
resulting Hall voltage falsifies the R
xxLo
measurement. To eliminate this unwanted
current, small inphase and quadrature voltages are injected into the low junction of
the quantum Hall resistance until the voltage at the port B, U
B
, becomes zero. As a
result, the capacitive current in lead B is zero; it is not exactly zero in lead A but so
small that no significant inphase voltage drop occurs along lead A. Therefore, the
contact and lead resistances of leads A and B have no significant effect on the R
xxLo
measurement.
Y
3Lo
I
3Lo
I
3Lo
R
H
I
2Lo
R
H
I
2Lo
I
1Lo
Z
1Lo
Y
1Lo
Y
2Lo
Z
2Lo
R
xxLo
Z
3Lo
C
inj
I
inj
Low
bU aU
A B
U
2DEG
D
U
C
int
+ C
ext
R
H
U
A
U
B
=
0
Figure 9.14 Equivalent circuit of a R
xx
measurement at the lowpotential side of a
quadrupleseries quantum Hall device
274 Coaxial electrical circuits for interferencefree measurements
Also, the capacitance between the Aterminal and the highpotential side of the
quantum Hall device drives a current through the Aterminal. This capacitance
consists of an internal contribution, C
int
, through the twodimensional electron gas
and an external stray capacitance to highpotential conductors, C
ext
. In contrast, the
capacitance between the Aterminal and conductors connected to the shield, which
are close to the device, have no significant effect because the voltage across these
capacitances is practically zero. As a first step, a capacitive current I
inj
is injected into
the Aterminal by means of a lowloss capacitance standard C
inj
(Figure 9.14); 1 pF is
a suitable capacitance value. Then the voltage at the port A, U
A
, is measured. The
injected quadrature current is adjusted until the detector signal is minimal (i.e. until
the quadrature component due to the parallel capacitances is compensated). The
phase of the lockin amplifier is then adjusted so that an imbalance of the injection
capacitance C
inj
is precisely in quadrature. R
xxLo
is now calculated from
U
A
I
R
xxLo
jwR
2
H
C
int
1 j tan d
int
C
ext
1 j tan d
ext
C
inj
1 j tan d
inj
9:18
The first three terms including the internal and external capacitances and
associated loss factors tan d
inj
and tan d
ext
, respectively, are the required AC
longitudinal resistance. The term tan d
inj
describes the phase error of the injected
quadrature current arising from the loss factor of the injection capacitor or a phase
error of the voltage applied. The measurement is sensitive to phase errors because
the prefactor wR
H
2
in (9.18) is large, even if the internal and external capacitances
are very small. The quadrature component of the longitudinal voltage is easily
observed, being about three orders of magnitude greater than the resolution of the
measurement. If the loss factors are well below 10
3
, the associated inphase
components are negligible.
If a quantum Hall device is bonded for a quadrupleseries connection scheme,
the longitudinal resistance can be measured with AC at both pairs of lowpotential
terminals. The potential of the unused low connection is extremely close to zero,
and so it makes no significant difference if a cable is connected to the low junction
of the device or it simply remains open.
Figure 9.15 shows the circuit of a practical R
xxLo
bridge. The phasesensitive
detector follows a singleended preamplifier. Instead of the capacitor C
inj
being
adjustable, a capacitor with a fixed value is supplied from the adjustable output
voltage of an IVD. This IVD is not directly connected to the sine generator but to
the high junction of the quantum Hall resistance to avoid phase errors, as men
tioned above. The supply transformer provides isolation and drives gates or shields,
which are not explicitly shown.
The potential at port B may change when a field sweep is carried out (without
rebalancing the injection system), or it might be nonzero due to an imperfect
balance or due to drift effects. To greatly reduce the resulting error, lead B can be
Bridges to measure impedance ratios 275
shortcircuited after nulling the potential at port B. A switch box can be used for
this purpose and for changing the detector between ports A and B.
In general, when a nonzero voltage, for example, the longitudinal voltage, is
measured at the output port of a bridge and the detector or the detector lead has an
input impedance that is comparable with the output impedance of the bridge, the
bridge is loaded and the displayed voltage is smaller than the voltage of the unloaded
bridge. The loading factor L is the ratio of the observed voltage to that which would
be measured if there were no loading. It can be calculated and is given by
L
1
1 R
out
=R
det
2
wR
H
C
det
2
q 9:19
R
out
( R
H
) and R
det
are the output impedance of the bridge and the input
impedance of the detector, respectively. C
det
is the parallel capacitance of the
detector lead. The loading function L decreases with frequency and can be either
calculated or directly measured when the bridge is imbalanced by a known amount.
For small longitudinal voltages, this loading correction is usually negligible and
thus not explicitly quoted in (9.18).
9.3.4 Measuring R
xxHi
The equivalent circuit for the AC R
xx
measurement at the highpotential side of a
quantum Hall device is shown in Figure 9.16. The lowpotential side of the
equivalent circuit is the same as shown in Figure 6.42 and thus not explicitly shown
in Figure 9.16.
B
D
0
+U
U
reference signal
generator
supply transformer high
low
A
B
C
inj
switch box
Figure 9.15 R
xx
measurement at the lowpotential side of a quantum Hall device
276 Coaxial electrical circuits for interferencefree measurements
Z
1Hi
Z
2Hi
Y
2Hi
Y
3Hi
Z
3Hi
Y
1Hi
I
1Hi
I
2Hi
I
3Hi
I
inj
C
int
+ C
ext
C
inj
C
g
I
off
C
off
R
off
I
2Hi
R
H
I
3Hi
R
H
R
xxHi
R
C1Hi
R
C2Hi
R
C3Hi
A B
U
2DEG
U
0
main
aux
U
0
2U
D
+
R
H
U
0
U
g
= 0
Figure 9.16 Equivalent circuit of a R
xx
measurement at the highpotential side of
a quadrupleseries quantum Hall device
The main difference with respect to the R
xxLo
measurement is that the A and
Bterminals are now at high potential and cannot be balanced to zero as can be done
at the lowpotential side. Therefore, the admittance Y
4Hi
of lead A draws a large
current I
4Hi
Y
4Hi
U through the Acontact of the quantum Hall device. The cor
responding Hall voltage I
4Hi
R
H
of that current amounts to about 0.1U and com
pletely swamps the longitudinal voltage IR
xxHi
. It is practically impossible to have a
current detector sufficiently close to the Acontact, so I
4Hi
cannot be directly
measured and nulled by any current injection system. Several other but smaller
voltages are also superposed on the longitudinal voltage (e.g. the commonmode
rejection signal of the differential preamplifier). All these unwanted voltages have
the common property that they do not directly depend on the Hall current I and
can be combined as one offset voltage, U
off
. The details of the individual con
tributions will turn out not to be relevant, and they are thus not listed here in detail.
Analysing the equivalent circuit of Figure 9.16 yields for the differential voltage
detector
U
Det
IR
xxHi
U
off
IR
H
P
jIwR
2
H
C
int
1 j tand
int
C
ext
1 j tand
ext
C
inj
1 j tand
inj
9:20
with U
off
UY
4Hi
R
H
I
off
R
H
higherorder terms
and P
Y
n 1
k1
~
Z
kHi
R
H
9:21
The compensation of the offset voltage can be achieved by injecting a current
I
off
through a capacitor C
off
and a resistor R
off
into lead A. To distinguish between
the longitudinal voltage IR
xxHi
and the offset voltage U
off
, the following procedure
Bridges to measure impedance ratios 277
is employed: the low junction of the multipleseries connection is temporarily
switched from low to high potential. This configuration is called the auxiliary
mode. As a result, the Hall current I flowing through the quantum Hall device (and
thereby the corresponding longitudinal voltage) is switched off, while the offset
voltage remains the same. In this configuration, the offset compensation system is
balanced until the detector is nulled.
I 0 ) U
Det
U
off
0 9:22
When the low junction of the multipleseries connection is now reconnected
to low potential (main mode), the offset voltage U
off
remains zero and the
detector voltage divided by the main Hall current I gives the AC longitudinal
resistance.
U
Det
I R
H
P
R
xxHi
jwR
2
H
C
int
1 j tan d
int
C
ext
1 j tan d
ext
C
inj
1 j tan d
inj
9:23
To compensate for the internal and external capacitances of the quantum Hall
device, a capacitive current I
inj
is injected into terminal A via a highquality
capacitance C
inj
(Figure 9.16). This injection system is set up in such a way that the
capacitive current is injected only in the main mode (and not in the auxiliary mode)
and is adjusted until the detector signal is minimal. The phase of the lockin
D
high
A B
0
+U
U
reference signal
generator
supply transformer
B
low
main
auxiliary
C
off
C
inj
R
off
offset compensation IVD
+
Figure 9.17 Circuit diagram of the AC R
xxHi
bridge
278 Coaxial electrical circuits for interferencefree measurements
amplifier is adjusted so that an imbalance of C
inj
occurs precisely in quadrature.
Then only the inphase component remains, while the quadrature component, if
desired, can be determined from the value of C
inj
and the setting of the particular IVD.
Equation (9.23) is formally identical with (9.18) for the lowpotential side of
the AC quantum Hall resistance, apart from the term R
H
P on the lefthand side
(defined in (9.21); see also (6.48)). This correction term occurs because lead B is
not directly connected to the respective contact of the quantum Hall device but to
the high junction of the multipleseries connection scheme. The potentials at both
points are slightly different because the small residual current I
3Hi
PI causes a
small voltage drop along the penultimate highpotential lead. In the case of a
quadrupleseries connection (n 4), P is typically smaller than 10
10
and there
fore negligible. In the case of a tripleseries connection (n 3), P has a typical
magnitude of a few parts in 10
8
. It is therefore necessary to apply a small correc
tion, which can be calculated with sufficient accuracy from the measured con
necting impedances
~
Z
k
(see section 6.7.2).
Figure 9.17 shows the scheme of a practical R
xxHi
bridge for a quantum Hall
device in tripleseries connection scheme. The supply transformer provides isola
tion and drives gates or shields, which are not explicitly shown. To ensure that the
voltage driving the offset compensation system does not change when altering
between the two configurations, the input terminal of the corresponding IVD is
connected to the high junction of the multipleseries connection and not directly to
the generator or to the supply transformer. Direct connection would cause sig
nificant loading effects.
Replugging the relevant leads to set the low junction of the multipleseries
connection temporarily to a high potential would give rise to a significant sys
tematic error. Instead, a switch is used so that the impedance of the outer con
ductors and of the whole R
xxHi
bridge remains the same for both configurations.
However, the bridge output impedance is not exactly the same in the two modes. In
the auxiliary mode, the inphase current through the quantum Hall resistance is
switched off, while at the same time, an extra capacitive current is switched on in
those particular lowpotential leads, which are temporarily at a high potential. This
loads the commonmode rejection of the detector and may cause a significant error.
It can be avoided by adding a matched capacitor and a 12.9kW resistor into the
switch box, as shown in Figure 9.17, to ensure that the bridge impedance remains
the same for both configurations.
The value of the required offset compensation capacitance C
off
is of the order
of 10 nF. Its impedance at a frequency of 1 kHz is comparable to R
H
and further
decreases with frequency. This would seriously load the R
xxHi
bridge and falsely
reduce the detector signal (see (9.19) in section 9.3.3). To avoid this loading effect,
the IVD driving the offset capacitor can be centre tapped so that the voltage across
the offset capacitor is larger and a correspondingly smaller capacitance can be used,
for example, C
off
400 pF. Also, the capacitance of the detector lead loads the
R
xxHi
bridge, and thus, it should be kept small. By these measures, the loading
effect can be kept reasonably small and, if necessary, a small correction can be
Bridges to measure impedance ratios 279
applied ((9.19) in section 9.3.3). Because the resolution of the R
xxHi
bridge is about
seven orders of magnitude greater than the compensated capacitance of lead A, the
bridge is sensitive to very small fluctuations and drift of the combined capacitance
of lead A and the offset compensation capacitor. Therefore, an offset capacitor with
a good shortterm stability is required, and lead A should not be disturbed during
the measurement.
The required value of the offset compensation resistor R
off
is of the order of
20 MW. Such a resistor is found to show unwanted fluctuations and generates a
high thermal noise power, and its parallel capacitance shows fluctuations. A better
stability and less thermal noise have been achieved by replacing the 20MW resistor
by a 200kW resistor. Then more dials of the corresponding IVD have to be
balanced (six instead of four), and so the improvement in stability comes along
with a loss of sensitivity. If the IVD driving the offset resistor is provided with an
internal stepup transformer (with a ratio of e.g. 50), this compensates for the loss
of sensitivity and improves the performance of the R
xxHi
bridge.
The previous paragraphs describe how to measure R
xxHi
at the last pair of high
potential terminals. If a quantum Hall device is bonded in quadruple series, R
xxHi
can also be measured at the penultimate pair of highpotential terminals. The last,
unused highpotential lead can either be connected to the high junction of the
multipleseries connection or simply remain open.
9.3.5 A simple coaxial bridge for measuring nondecade
capacitances
Commercial capacitance bridges may yield inaccurate results when trying to measure
noncoaxially threeterminal capacitances through long leads to the quantum Hall
device in a cryostat. The simple coaxial bridge suggested as an alternative below is a
very useful tool that can carry out an accurate, lownoise measurement even at a very
low voltage level.
The tool shown in Figure 9.18, a simple coaxial twoterminalpair ratio bridge
for measuring nondecade capacitances and their associated loss factors, is a very
useful tool for investigating quantum Hall devices. An example of an application is
the measurement of the capacitance between the twodimensional electron gas of a
device and a back or side gate. This capacitance can be measured, for example, as a
function of the magnetic field or as a function of frequency and current. Another
application is the measurement of the loss factor of different materials, which are
selected for mounting a device.
The unknown capacitance is measured in terms of the capacitance of an air
dielectric lowloss rotary vane capacitor whose maximum value can be chosen to
be, for example, either 1 pF or 10 pF and whose minimum value is 50 fF. A ratio
transformer, which can be configured for different ratios like 1:1, 10:1 or 1:10,
allows a capacitance range from 5 fF to 100 pF to be continuously covered. The
phase of the detector is adjusted such that an imbalance of the rotary vane capacitor
causes a completely inphase response. The loss factor of the unknown capacitance
is then calculated from the quadrature component of the detector signal. Assuming
280 Coaxial electrical circuits for interferencefree measurements
that the rotary vane capacitor has a sufficiently low loss factor and the ratio
transformer has a sufficiently small phase error, this simple bridge yields an
uncertainty of less than 1 10
5
for the loss factor, even at a low voltage level of
0.1 V. If necessary, a Wagner balance can be added.
9.3.6 Coaxial resistance ratio bridges involving quantum
Hall devices
Quantum Hall resistance has the advantage that its quantised value depends only on
unchangeable fundamental constants and is precisely independent of time, tem
perature, barometric pressure and all the other side effects that conventional
resistance artefacts suffer from. Another advantage is the very low thermal noise
power of the cryogenic quantum Hall device compared to room temperature
resistance standards. Furthermore, the currents flowing through the defining Hall
potential contacts are automatically minimised by the multipleseries connection
scheme, and they are even smaller than can be manually adjusted in a conventional
fourterminalpair bridge. Therefore, a quantum Hall device in multipleseries
connection can automatically take over the function of both a current source and a
Kelvin arm. This property simplifies conventional fourterminalpair bridges (see
sections 9.1 and 9.2) as shown in the following examples.
Figure 9.19 shows a ratio bridge for comparing a tripleseries connected
quantum Hall device with a conventional fourterminalpair resistance standard.
While a ratio bridge for comparing two conventional fourterminalpair standards
requires a Kelvin network and two current sources with their associated current
detection transformers, the very low selfadjusted residual current in the defining
potential leads of the quantum Hall device is so small that the Kelvin network and
one current source can be omitted.
If a ratio bridge is to compare two quantum Hall resistances directly with each
other, the remaining current source and the associated current detection transformer
can be omitted (Figure 9.20). Not only this further simplifies the ratio bridge but
D
C
Ref
0
+NU
U
reference signal generator
backgate
2DEG
B
backgate
2DEG
C
X
(1 jD
X
)
Figure 9.18 A coaxial twoterminalpair ratio bridge for the comparison of non
decade capacitances
Bridges to measure impedance ratios 281
also, because both quantum Hall devices are operated at low temperature and no
other room temperature resistors are required in this bridge network, the noise
level is very low and a very high resolution is possible. To simplify the drawing,
Figure 9.20 shows the two quantum Hall devices as if they would experience a
magnetic field of different polarity.
9.3.7 A quadrature bridge with two quantum Hall devices
Multipleseries connected quantum Hall devices can also be directly incorporated
in a quadrature bridge ([26] of chapter 6). This simplifies the bridge network in the
same way as for the ratio bridge described in the previous section. Whereas a
conventional fourterminalpair quadrature bridge is the most complicated coaxial
optical bus fibre
PC
R
ref
R
H
Wagner
injection IVD
to detector
t
o
d
e
t
e
c
t
o
r
generator
D
reference signal
detector port
ratio transformer
current source
C
Q
0
B
supply
transformer
+U
U
Figure 9.19 Comparing a tripleseries connected quantum Hall device, R
H1
, with
a fourterminalpair resistor, R
ref
PC
D
0
B
B
optical bus fibre
R
H1
R
H2
Wagner
injection
IVD
t
o
d
e
t
e
c
t
o
r
generator
reference signal
detector port
ratio transformer
C
Q
supply
transformer
+U
U
Figure 9.20 Comparing two tripleseries connected quantum Hall devices, R
H1
and R
H2
282 Coaxial electrical circuits for interferencefree measurements
bridge network usually encountered (Figure 9.9 and section 9.2.3), requires a some
what tedious balancing procedure and suffers from a poor signaltonoise ratio,
the remarkable properties of quantum Hall resistance can greatly improve all these
aspects. Three combining networks and four injection/detection transformers can be
omitted, as shown in Figure 9.21. As a result, the balancing procedure and convergence
are also improved. Finally, cryogenic quantum Hall resistances permit a very low
overall thermal noise level if the resistors of the detector combining network are also at
a low temperature and if a lownoise harmonic filter (see Figure 2.4 and section 2.1.2)
and a lownoise preamplifier (see section 2.3.1) are used. Altogether, the operation of
such a quadrature bridge is, compared to a full fourterminalpair quadrature bridge
with room temperature resistance standards, amazingly simple.
9.4 Highfrequency networks
Callegaro [8] has published a useful review, containing an exhaustive list of
references, of this subject.
The networks in this section are modelled closely on those designed for kilo
hertz frequencies described in sections 9.1 and 9.2. The main variation is that the
components are smaller and the lengths of interconnecting cables shorter to main
tain the dimensions of the network much less than a quarter wavelength at the
operating frequency. Because only more modest accuracies are needed at higher
frequencies, some simplifications are possible to help fulfil this need. In particular,
current equalisers where a cable is threaded only once through a core are usually
adequate for frequencies up to 1 MHz. At frequencies greater than about 1 MHz,
they can be omitted entirely because the goandreturn inductance of a cable is
about a factor of 10 less than the loop inductance and resistance of its outer con
ductor even if the loop inductance is not enhanced by threading a core.
Because of their small size, SubMiniature type B (SMB) connectors are
suitable for cable terminations and the ports of components, but they are somewhat
delicate. To conserve their working life, a suitable tool should be constructed and
used so that only axial force is applied to plug and unplug them.
Small magnetic cores wound from nanocrystalline metal ribbon of micrometre
thickness have sufficient permeability for frequencies up to 1 MHz to construct
transformers described in this section. Highpermeability ferrite cores are needed
for higher frequencies. In general, transformers can be kept below selfresonance
with sufficient input impedance by using far fewer turns than at kilohertz fre
quencies. This will also minimise the parasitic inductances, resistances and inter
winding capacitances, which cause departure from their nominal voltage ratio.
9.4.1 An IVDbased bridge for comparing 10:1 ratios of impedance
from 10 kHz to 1 MHz
Figure 9.22 is a diagram of this bridge. The highpotential ports of the four
terminalpair impedances Z
A
and Z
B
are connected with identical cables to the ratio
windings of the twostaged inphase ratio IVD. The highcurrent ports of Z
A
and Z
B
Bridges to measure impedance ratios 283
D
r e f e r e n c e
f
i
l
t
e
r
+
j
U
0
B
B
o
p
t
i
c
a
l
b
u
s
f
i
b
r
e
P
C
R
H
1
R
T
1
R
T
2
R
H
2
C
T
2
C
T
1
W
a
g
n
e
r
i
n
j
e
c
t
i
o
n
I
V
D
t
o
d
e
t
e
c
t
o
r
t
o
d
e
t
e
c
t
o
r
g
e
n
e
r
a
t
o
r
d
e
t
e
c
t
o
r
p
o
r
t
r
a
t
i
o
t
r
a
n
s
f
o
r
m
e
r
c
u
r
r
e
n
t
s
o
u
r
c
e
C
Q
C
1
C
2
C
P
C
J
R
J
s
u
p
p
l
y
t
r
a
n
s
f
o
r
m
e
r
+
U
U
F
i
g
u
r
e
9
.
2
1
A
q
u
a
d
r
a
t
u
r
e
b
r
i
d
g
e
,
w
h
i
c
h
l
i
n
k
s
t
w
o
t
r
i
p
l
e

s
e
r
i
e
s
c
o
n
n
e
c
t
e
d
q
u
a
n
t
u
m
H
a
l
l
r
e
s
i
s
t
a
n
c
e
s
,
R
H
1
a
n
d
R
H
2
,
a
n
d
t
w
o
c
a
p
a
c
i
t
a
n
c
e
s
,
C
1
a
n
d
C
2
,
a
c
c
o
r
d
i
n
g
t
o
w
2
R
H
1
R
H
2
C
1
C
2
1
284 Coaxial electrical circuits for interferencefree measurements
are connected to its magnetic winding to minimise the current in the potential con
nections. The defining condition of zero current at the IVD ports is determined by
extrapolation of the results obtained from altering the lengths of these cables. The
adjustable impedances z
a
and z
b
at the lowpotential ports of Z
A
and Z
B
are small
510 pF trimmer capacitors, which form combining network. The lowcurrent ports
of the standards are connected via a 500W resistance R, which can be shorted by a
switch S. S is opened for balancing the combining network and closed to balance the
bridge. These balances are iterated as necessary. The condition of zero current
through the potential reference point O of the bridge is obtained by adjusting the
Wagner capacitance C
W
so that the detector is nulled when the shorting plug at O is
removed.
If Z
A
and Z
B
are capacitive, 1/Y should be a resistance, and the bridge balance
equations are
C
A
C
B
n
1 n
9:24
where n is the ratio IVD setting
and
D
A
D
B
YwC
B
1
k n
n
9:25
where k is the injection IVD setting.
An accuracy of 0.01% has been achieved when measuring the ratio of 100pF
and 1nF gasdielectric capacitors.
A general method for calibrating the IVD voltage ratios similar to sections
7.4.1 to 7.4.5 has yet to be developed, but particular ratios including 10:1 can be
calibrated using the permuting capacitors device [9], described in section 7.4.6 in
the simple bridge drawn in Figure 9.23. Alternatively, a ratio transformer in a
D
supply
transformer
reference signal
generator
ratio IVD
Y
injection IVD
C
W
Kelvin arm
R
z
A
z
B
S
Z
B
0
n1
n
k
Z
A
Figure 9.22 A quasifourterminalpair IVDbased bridge for frequencies up
to 1 MHz
Bridges to measure impedance ratios 285
bridge can be calibrated by balancing the bridge with the 0.1pF and 1pF or the
1pF and 10pF adjustable capacitors described in section 6.5.8. The transformer
ratio at any frequency can then be inferred from their values obtained by measuring
their 1kHz values and assuming their frequency dependence as calculated from their
selfresonant frequencies. This method will, in general, be less accurate than the
permuting capacitors method, but it may be adequate and is quicker to implement.
9.4.2 A bridge for measuring impedance from 10 kHz to 1 MHz
based on a 10: 1 voltage ratio transformer
This network (Figure 9.24) is similar in construction to that of the previous section.
The ratio IVD is replaced by a 10:1 fixed ratio voltage transformer, and current
detection transformers have been added to the voltage ratio ports of the transfor
mer. Detectors connected to the current detection transformers are nulled to ensure
zero current through these ratio ports by adjusting the binary current sources 1 and
2 connected to the highcurrent connections of Z
1
and Z
2
. The twostaged high
frequency injection IVD provides the main bridge balance via appropriate admit
tances Y
A
and Y
B
(inphase and quadrature injection components). This IVD as well
as the main 10:1 voltage ratio transformer are calibrated at frequencies up to 1 MHz
by the highfrequency permuting capacitors technique (see section 7.4.6). As the
accuracy of this complete 4TP bridge for capacitance or resistance standards is only
10 ppm at 1 MHz, current equalisers are not necessary, and the small frequency
dependences of the injection admittances Y
A
and Y
B
have negligible effect.
9.4.3 Quasifourterminalpair 1:1 and 10:1 ratio bridges for
comparing similar impedances from 0.5 to 10 MHz
Figure 9.25 is the circuit diagram of a network that can be configured with an
appropriate ratio transformer as a 1:1 or 10:1 bridge. The combining network is two
coaxial cables whose lengths are also in the same ratio. This bridge is very
D
reference signal
ratio IVD
Y
0
C
W
supply
transformer
generator
injection IVD
Y
10
Y
1
Figure 9.23 Calibrating a 10:1 ratio transformer or IVD with a permuting
capacitors device
286 Coaxial electrical circuits for interferencefree measurements
convenient for calibrating commercial highfrequency capacitance standards, par
ticularly of the type illustrated in the lower half of Figure 9.29.
1:1 and 10:1 twostaged ratio transformers can be constructed using
highfrequency ferrite cores, which exhibit low reluctance and small loss over the
0.510 MHz frequency range. They remain below selfresonance if both stages are
wound with about 10 spaced turns of 0.3mmdiameter polytetraflourideethelene
to generator
D
reference signal
binary current source 1
binary current source 2
generator
supply
transformer
injection IVD
detector port
to detector
to detector
to detector
ratio
transformer
z
A
z
B
Y
B
Y
A
Z
1
Z
2
Kelvin arm
Figure 9.24 A complete 4TP 1MHz impedance bridge based on a fixed
10:1 voltage ratio transformer and binary IVD current sources
D
reference signal
0
generator
supply transformer
ratio transformer
Y
10
C
W
Y
1
Figure 9.25 A bridge for comparing similar impedances from 0.5 to 10 MHZ
Bridges to measure impedance ratios 287
(PTFE) insulated wire, and their impedance does not unduly load a 50W source.
For the 1:1 transformer, the first core has ten turns, the second core is placed axially
next to it and the combination wound with ten turns, interleaved with the exposed
portions of the firststage winding so that turns of similar potential are adjacent, and
tapped after five turns. The ends and tap of the secondstage winding terminated on
three SMB connectors soldered into one end cap of the 22mmdiameter, 25mm
long copper cylindrical enclosure. The first stage terminates on two SMB con
nectors soldered into a small subplate and bolted to but insulated from the other
end cap.
The 10:1 ratio transformer is broadly similar, except that its firststage winding
has 11 turns. The secondstage winding, a single axial conductor between con
nectors soldered into the centre of each end cap constitutes the 1 winding when a
shorting plug is inserted into the connector having the appropriate end of the ten
turn winding connected to its inner pin. The shorting plug can be removed to adjust
C
W
to make the Wagner balance.
9.4.4 A fourterminalpair 10MHz 1:1 resistance ratio bridge
The circuit diagram is shown in Figure 9.26. The circuit again follows low
frequency coaxial bridge principles. The isolation transformer and the provision of
the 1:1 voltage ratio are combined into a single transformer. This reduces the
complexity of the bridge significantly, and this is important for accurate high
frequency measurements in order to keep the circuit simple and therefore the
interconnecting cables short. The capacitance between the primary and secondary
windings of the isolation transformer is 0.8 fF. Z
1
and Z
1
are the main four
terminalpair resistance standards, and their highcurrent terminals are driven from
the 2U and 2U potential terminals of the isolation/ratio transformer via adjus
table binary parallel capacitive and resistive components having 5bit coarse
resolution. The highpotential terminals of the standards are connected to the U
and U terminals of the isolation/ratio transformer via detection transformers D
1
and D
2
. The 0 tap of the transformer is the bridge zero defining point, being the
only point where its inner and outer conductors are shorted together. The low
current and lowpotential terminals of the standards are connected to the switch S
and a combining network consisting of a 100W miniature variable resistance z
A
and a fixed 50W resistance z
B
. The bridge balance procedure is the same as that of
a similar lowfrequency bridge except that the bridge inphase and quadrature
balance condition is calculated by extrapolation from the small change produced in
the detector reading by shunting a known impedance DZ across one of the standards
and readjusting the auxiliary balances. The linearity of the detector is verified
separately, and the error from the frequency dependence of DZ can be sufficiently
small. Typically, if Z
1
and Z
2
are nominally 1 kW, DZ could be between 10 and
100 kW. The error of the 1:1 transformer ratio is eliminated by interchanging the
connections to it and taking the mean of the results.
The complete bridge system is contained within a 0.25m radius. This is suf
ficiently smaller than the 7.5m quarter wavelength at 10 MHz that all the bridge
288 Coaxial electrical circuits for interferencefree measurements
components can be modelled by lumped parameters. Also, reflections caused by
differing characteristic impedances between the cables, connectors and standards
are not significant. The correct functioning of the bridge can be tested by com
paring two identical highfrequency calculable coaxial resistance standards (see
section 6.5.3).
9.4.5 A 1.6 and 16MHz quadrature bridge
A quasifourterminalpair quadrature bridge system operating at 1.592 MHz is
shown in Figure 9.27. The bridge operates in a similar manner to the conventional
lowfrequency quadrature bridge system, discussed in more detail in sections 9.2.2
and 9.2.3. Measurements confirming correct operation have been made by directly
comparing the calculated and measured frequency dependence of two 100pF
capacitance and two 1kW resistance standards between 1.592 kHz and 1.592 MHz
(10
4
rad/s and 10
7
rad/s, respectively). Operation at 15.92 MHz (10
8
rad/s) is also
possible using the same bridge system as well as the same capacitance standards
but with two 100W resistance standards. The network in Figure 9.27 compares the
mean value of capacitors C
1
and C
2
to that of resistors R
1
and R
2
. On the right is a
D
r
e
f
e
r
e
n
c
e
s
i
g
n
a
l
detector port
+2U
0
+U
U
2U
Z
1
Z
1
Kelvin arm
Z
K
z
A
z
B
800 100 120
90
50
800 100 120 50
50 pF 100 pF 800 pF
365 pF
50 pF 100 pF 800 pF
365 pF
Z
isolation transformer
generator
to detector
to detector
to detector
Figure 9.26 A 1:1 fourterminalpair 10 MHz resistance ratio bridge
Bridges to measure impedance ratios 289
highfrequency supply transformer having output voltages of U and 2U. The
ratio of the potentials at the P
H
ports of R
1
and C
2
is measured by the subbridge
consisting of 1pF capacitors. These parallelplate airdielectric capacitors are
designed for minimum change of value with frequency and are of as identical
construction as possible, so the residual change with frequency is the same for each
capacitor. They are provided with conducting screws, which impinge on their
fringing electric fields, thus enabling their value to be finely adjusted (see section
6.5.8). Although provided with fourterminalpair terminations, here only their H
P
and L
C
ports are connected so that they become twoterminalpair components. A
highfrequency phasesensitive detector is temporarily connected to them and
balanced by adjusting one or the other capacitor slightly, and the ratio of their
values at 1 kHz is measured on a separate highresolution bridge by carefully dis
connecting them temporarily from the network. When reconnected, this ratio is also
the required ratio of the potentials at the P
H
ports of C
1
and R
4
. The small and
nearly equal input impedances of these capacitors (~3 10
4
W and ~3 10
3
W at
1.6 and 16 MHz, respectively), together with that of any identical cables, ensure
that the departure from I 0 defining condition at these ports is violated only to a
small and nearly equal extent.
The jnetwork is connected to the 2U ports of the supply transformer by two
short cables. It consists of C
J
and R
J
. The capacitor C
J
is ~68 pF in parallel with a
565 pF miniature vane trimmer capacitor, housed in a small copper box. The
resistor R
J
is either a 560W resistor in series with a 01 kW miniature trimmer
resistor or a 56W resistor in series with a 0100 W miniature trimmer resistor,
housed in another copper box. These resistor combinations can be switched into the
circuit with a miniature toggle switch for bridge operation at 1.6 or 16 MHz,
r
e
f
e
r
e
n
c
e
s
i
g
n
a
l
0
1 pF
1 pF
+jU
R
T1
R
1
R
T2
R
2
C
T2
C
T1
to detector
C
B
A
to detector
generator
detector port
D
C
1
C
2
C
J
R
J
supply transformer
+U
+2U
U
2U
Figure 9.27 A quasi fourterminalpair quadrature bridge operating at 1.6 or
16 MHz
290 Coaxial electrical circuits for interferencefree measurements
respectively. These components, and those in the detector combining network
described below, have been verified on an impedance analyser to have adequate
highfrequency performance, and since they form an auxiliary network adjusted to
a balanced state, their exact values are immaterial. An insulated trimmer tool can
be inserted through a small hole in their housing boxes to adjust their values. The
components are mounted within the box sufficiently far from the aperture to ensure
that the effects of emanating electric fields on the rest of the bridge network
are negligible.
The detector combining network is similarly housed in a copper shielding box.
Either a 1kW or a fixed 100W resistor R
T2
is connected in the circuit by a switch
for 1.6 or 16MHz operation, respectively, and the 1kW or 100W adjustable
resistors R
T1
, similarly by another switch. The detector combining network is
adjusted by altering the appropriate variable resistor and the variable capacitor to
null the detector. The bridge network also contains three nonadjustable combining
networks, A, B and C made from lowimpedance coupling connectors joining the
defining ports P
H
or C
L
and other connectors joining the P
L
or C
H
ports adjacent
main bridge components in such a way that the potential or current pickoff points
are in their centres. There are no current equalisers as the goandreturn impedance
of the coupling connectors and cables is more than a factor of 10 lower than the
loop impedances of their outer conductors, thus ensuring that current balancing in
the interconnecting cables is adequate for the required level of accuracy of about
0.1% at 1.6 MHz.
On comparing the known values of R
1
, R
2
, C
1
and C
2
with those measured on
the quasifourterminalpair HF quadrature bridge, Awan and Kibble [10] found a
difference of only 0.17% at 1.6 MHz.
9.4.6 Fourterminalpair resonance frequency measurement
of capacitors
The fourterminalpair resonance frequency f
0
of capacitance standards can be
determined by measuring with the circuit shown in Figure 9.28a a set of four
parameters over a range of frequencies near f
0
. The capacitances C
i
and C
d
are
chosen to be small enough (typically 0.1 pF) not to perturb the measurement
parameters significantly. These parameters are the twoterminalpair impedances
appearing in the impedance matrix (Z) defined in (9.26).
V
1
V
2
V
3
V
4
2
6
6
4
3
7
7
5
Z
11
Z
12
Z
13
Z
14
Z
21
Z
22
Z
23
Z
24
Z
31
Z
32
Z
33
Z
34
Z
41
Z
42
Z
43
Z
44
2
6
6
4
3
7
7
5
I
1
I
2
I
3
I
4
2
6
6
4
3
7
7
5
9:26
where Z
ij
V
i
/I
j
(i,j 1,2,3,4). For a linear network obeying reciprocity, Z
ij
Z
ji
.
The fourterminalpair impedance is, in the notation of Figure 9.28b,
Z
4TP
V
2
I
4
I
2
I
3
V
3
V
4
0
9:27
Bridges to measure impedance ratios 291
From knowledge of f
0
, the frequency dependence of the capacitor for fre
quencies f ( f
0
can be calculated from (6.43). It is also possible to calculate the
dissipation factor of gasdielectric capacitors from the 3 dB width of the mea
sured resonance curve, that is, the ratio of the difference of the frequencies f
1
and f
2
to f
0
, which are 3 dB down from the peak at f
0
. Capacitors whose frequency
dependence has been determined in this way can calibrate LCR meters and impe
dance analysers at frequencies that are sufficiently lower than f
0
.
Equation (9.27) can be rearranged using the singleport definitions given in
(9.26) to give [11]
Z
4TP
Z
24
Z
31
Z
21
Z
34
Z
31
9:28
9.4.7 Scattering parameter measurements and the link to
microwave measurements
The instruments commonly used for impedance measurements between 10 MHz
and about 300 GHz are two or fourport vector network analysers. These instru
ments can also make fourterminalpair measurements, Z
4TP
, as detailed below.
In the socalled indirect Zmatrix method [12,13], Z
4TP
is estimated by
measuring the individual elements Z
ij
of the 4 4 Zmatrix of (9.26) with a one
port network analyser. The diagonal elements Z
ii
are measured directly with the
other three ports left open. The offdiagonal elements Z
ij
are measured using the
relation
1
2
3
4
C
C
H
C
H
P
H
P
H
P
L
P
L
C
L
C
L
R R
1
L
1
C
i
C
2
C
4
C
d
C
D
R
D
L
3
R
3
I
3
U
2
= U
I
2
= 0 I
4
= 0
U
4
= 0
I
1
D
C
(a)
(b)
L
generator
Figure 9.28 (a) A circuit for measuring the fourterminalpair resonance
frequency of a gasdielectric capacitor C. (b) The fourterminalpair
definition of the capacitor
292 Coaxial electrical circuits for interferencefree measurements
Z
ij
Z
ji
Z
jj
Z
ii
Z
sj
ij
q
9:29
where Z
sj
ij
is the impedance measured at port i when port j is shorted and the
remaining ports are left open. By combining (9.28) and (9.29), we obtain the
Suzuki relation for a fourterminalpair admittance in terms of singleport
definitions:
Y
4TP
Z
11
Z
s3
11
Z
22
s
Z
11
Z
s2
11
Z
44
Z
s3
44
q
Z
11
Z
s3
11
Z
44
Z
s2
44
q
!
1
9:30
and hence, Z
4TP
(Y
4TP
)
1
. The limitation of this method is that oneport mea
surements made at a given frequency with network analysers are not sufficiently
accurate for an evaluation of Z
4TP
, due to, for example, contact impedances, cable
loading and matching, and instrument drift. These difficulties have been overcome
by making measurements over a much larger bandwidth than the frequency range
of interest to identify the magnitude of the residual and parasitic impedances. To
date, this method has only been applied to a limited range of gasdielectric capa
citance standards.
Callegaro and Durbiano [14] showed that it is also possible to measure four
terminalpair impedance standards using scattering parameters measured with a
vector network analyser. Since an impedance matrix [Z] and its scattering para
meter matrix [S] are related by
S Z Z
0
IZ Z
0
I
1
9:31
where [I] is the identity matrix of order n, they showed that a fourterminalpair
impedance can be written as
Z
4TP
2Z
0
s
21
s
34
s
31
s
24
s
31
s
21
s
32
s
31
s
44
s
31
s
22
s
41
s
34
s
21
s
32
s
44
s
21
s
34
s
42
s
31
s
22
s
44
s
31
s
42
s
24
s
41
s
34
s
22
s
41
s
24
s
32
!
1
9:32
where Z
0
is the characteristic impedance of the network. If the network is termi
nated with a cable of characteristic impedance Z
0
H(Z/Y), then the cable appears
to the source to be infinitely long and other sections of the same characteristic
impedance can be added without affecting the source. In contrast, with guided
waves, a transverse electromagnetic wave in free space has a characteristic impe
dance of H(e
0
/m
0
) 377 W.
Although there are some advantages of performing sparameter measurements
to determine a fourterminalpair impedance, the method is generally only accurate
for impedance values close to Z
0
. In this case, since s
ij

2
1, most of the significant
Bridges to measure impedance ratios 293
terms in (9.32) are transmission sparameters (i.e. s
21
, s
24
, s
31
and s
34
). Reference
plane definitions and the effect of impedancematched connector adaptors are less
critical than in reflection measurements. Furthermore, Z
4TP
can be directly com
puted from (9.32) instead of relying on an equivalent electrical model of the device
being measured. A fourport vector network analyser can make the measurements
in a single run, but with a twoport vector network analyser, the matrix elements of
[S] have to be measured sequentially by connecting two terminal pairs at a time
while terminating the remaining two terminals with matched loads, as illustrated in
Figure 9.29.
T/R test set
m
a
t
c
h
e
d
l
o
a
d
m
a
t
c
h
e
d
l
o
a
d
Osc R A B
network analyser
calibration
plane
NBNC adapters
1
1 2 3 4
2
fourterminalpair
impedance standard
I
H
V
H
V
L
I
L
Figure 9.29 An example of a connection between a fourterminalpair impedance
to a twoport vector network analyser. The particular connection
scheme shown is to measure s
22
and s
42
For example, scattering parameter measurements of a General Radio 1nF
capacitance standard calculated from the above expressions shows that the apparent
capacitance of the standard is about 30% higher at 10 MHz compared to its 1 kHz
value (Figure 9.30b).
294 Coaxial electrical circuits for interferencefree measurements
Despite the convenience and elegance of this method, evaluation of the
uncertainty is difficult and is still the subject of debate by RF and microwave
experts. In particular, it is not clear how to evaluate the uncertainties of the input
quantities in (9.32) and how to address their propagation through this equation
because it involves complex functions and strong correlations. Callegaro [14] has
suggested performing a numerical uncertainty evaluation with propagation being
accounted for by the Monte Carlo method.
9.4.8 Electronic fourterminalpair impedancemeasuring
instruments
Commercial automated LCR and impedance analyser instruments are commer
cially available, which claim to offer moderate accuracies, operate in milliseconds
and cover a wide frequency range from 20 Hz to 110 MHz and a wide impedance
range from 1 mW to 1 GW. In these precision instruments, an attempt is also made
to implement the fourterminalpair definition of impedances Z to be measured in
order to eliminate the effect of contact resistances and stray capacitances and
mutual inductances in the connections to the impedance. The maximum AC voltage
that these instruments supply for impedance measurement is usually 1 V
rms
, but
they can also provide up to 40 V DC bias for various applications. The claimed
measurement uncertainty varies depending on the impedance being measured and
the measurement frequency but is typically between 0.1% over most of their range
up to 10% at the boundaries of the instruments measuring capability. Their
applications are wide ranging, from electrical and electronic component and
materials evaluation to chemical and biological process measurements in a number
of industries such as aerospace, biomedical, general research and development and
telecommunications.
Most employ an autobalancing method in which highinputimpedance
amplifiers and feedback techniques create a virtual potential reference point P
(earth) at the lowpotential port of the instrument, as shown in Figure 9.31. The
instruments usually have four selectable range resistors, R
ref
, across which a voltage
s
c
a
t
t
e
r
i
n
g
p
a
r
a
m
e
t
e
r
s
i
j
Re [s
21
]
Im [s
21
]
Re [s
22
]
Im [s
22
]
Re [s
23
]
Im [s
23
]
frequency (MHz) frequency (MHz)
1 2 3 4 5 6 7 8 9 10 0 0.1
0.6
0.4
0.2
0.0
0.2
0.4
0.6
1 10 100
1.00 nF
1.05 nF
1.10 nF
1.15 nF
1.20 nF
1.25 nF
1.30 nF
1.35 nF
c
a
p
a
c
i
t
a
n
c
e
(a) (b)
Figure 9.30 (a) Scattering parameter measurements of a 1nF capacitor. (b) The
frequency dependence calculated from these measurements
Bridges to measure impedance ratios 295
U
2
is detected to determine the current flowing through Z. Often an assumption is
made that the highfrequency value of these resistors is the same as their DC value,
which is not entirely true for accurate measurements. These, and other imperfec
tions such as inexact obeying of the fourterminalpair defining conditions, are
corrected by internal computer software to obtain the expected results for known
calibrating impedances. Unfortunately, the connections to calibrating impedances
may not mimic sufficiently exactly connections to other devices to be measured
and there are other effects such as shunt capacitive loading at the high and low
potential ports, inadequate input impedance of the voltage detectors U
1
and U
2
over
the whole frequency range, inequality of the currents in the inner and outer con
ductors of connecting coaxial cables and incorrect accounting for cable corrections
(see section 5.4).
In quality control applications where pass/fail measurements are performed on
passive components relatively quickly, instrument manufacturers supply a number
of probes and adaptors. A common example is Kelvin clips in which four coaxial
cables are connected to the impedance analyser at one end and their inners are
connected to two crocodile clips and their outers are shorted together at the other.
The two inner surfaces of each crocodile clip are insulated from each other, and one
makes the potential connection, and the other, the corresponding high or low
current connection. A similar connection scheme is employed in various four
terminalpair to fourterminal adaptors for measuring chip surfacemounted
components.
For measurements at frequencies higher that 100 MHz up to at least 110 GHz,
twoport 50W matched impedance or vector network analysers are usually
employed.
At the less expensive end of the market, there are numerous lowaccuracy two
terminal handheld LCR meters. These typically measure a wide range of impe
dances at one or two fixed frequencies, 120 Hz and 1 kHz being the most common,
but offer at best only about 1% accuracy.
References
1. Cutkosky R.D. Techniques for comparing four terminalpair admittance
standards. J. Res. NBS 1970;740:6378
U
2
U
1
R
ref Z
P
D
D
Figure 9.31 The autobalancing technique typically employed in commercial
instruments to measure a fourterminalpair impedance Z
296 Coaxial electrical circuits for interferencefree measurements
2. Kibble B.P., Rayner G.H. Coaxial AC Bridges. Bristol: Adam Hilger Ltd.;
1984. (Presently available from NPL, Teddington, TW11 0LW, U.K.)
3. Awan S.A., Jones R.G., Kibble B.P. Evaluation of coaxial bridge systems for
accurate determination of the SI farad from the DC quantum Hall effect.
Metrologia. 2003;40 (5):26470
4. Hanke R., Kolling A., Melcher J. Inductance calibration in the frequency
range 50 Hz to 1 MHz at PTB. CPEM 2002 Digest. 2002;18687
5. Cote M. A fourterminal coaxialpair MaxwellWien bridge for the mea
surement of selfinductance. IEEE Trans. 2009;IM58 No. 4:96266
6. Sedlacek R., Bohacek J. Bridges for calibrating four terminalpair standards
of selfinductance at frequencies up to 10 kHz. Meas. Sci. Technol. 2009;
20:25105
7. British Standards Institution BS10413. BSI 389 Chiswick High Road,
London, W4 4AL U.K.
8. Callegaro L. The metrology of electrical impedance at high frequency: a
review. Meas. Sci. Technol. 2009;20:114
9. Awan S.A., Kibble B.P., Robinson I.A. Calibration of IVDs at frequencies
up to 1MHz by permuting capacitors. IEE Proc. Sci. Meas. Technol.
2000;147 (4):19395
10. Awan S.A., Kibble B.P. A universal geometry for calculable frequency
response coefficient of LCR standards and new 10 MHz resistance and
1.6 MHz quadrature bridge systems. IEEE Trans. Instrum. Meas. 2007;56
(2):22125
11. Awan S.A., Callegaro L., Kibble B.P. Resonance frequency of fourterminal
pair airdielectric capacitance standards and closing the metrological impe
dance triangle. Meas. Sci. Technol. 2004;15:96972
12. Suzuki K. A new universal calibration method for fourterminalpair
admittance standard. IEEE Trans. Instrum. Meas. 1991;40 (2):42022
13. Koffman A.D., AvramovZamurovic S., Waltrip B.C., Oldham N.M.
Uncertainty analysis for four terminalpair capacitance and dissipation factor
characterization at 1 and 10 MHz. IEEE Trans. Instrum. Meas. 2000;49
(2):34648
14. Callegaro L., Durbiano F. Fourterminalpair impedances and scattering
parameters. Meas. Sci. Technol. 2003;14:17
Bridges to measure impedance ratios 297
Chapter 10
Application of interferencefree circuitry to
other measurements
10.1 Resistance thermometry (DC and lowfrequency AC)
Resistance thermometry is the most precise method of measuring temperature in the
range from 30 to 1000 K. The sensor is usually a compact resistor of platinum
(or platinumrhodium alloy for higher temperatures), whose value at ambient tem
peratures is about 25 or 100 W. The change of resistance with temperature is about
0.4% per Kelvin, so that measurement of the resistance to 1 ppm yields temperature
with a resolution of about 0.25 mK. A given thermometer can be calibrated at the
various fixed points, which are the melting or triple points of water and various
metals, and a temperature scale derived from interpolation between these.
10.1.1 DC resistance thermometry
Measuring the resistance of a thermometer by DC resistance comparison bridges is
the older technique and is still in use. A plethora of schemes of increasing com
plexity for connection to a Wheatstone bridge (see section 5.1.1) or a Kelvin double
bridge (see section 5.6.1) have been devised to overcome to a moreorless degree
the problem of the resistance of connecting wires affecting the measurement (see,
for example, Reference 4). We would advocate the simple and elegant use of just
two coaxial cables devised by Cutkosky and described in section 5.3.5. Because
one cable carries the measuring current to and from the thermometer and the other
connects its two potential contacts through which there is no current, both cables
are inherently currentbalanced with the outer conductors acting as screens, this
connection scheme is interference free. Any bridge or instrument that satisfies the
defining conditions of a fourterminal resistance will yield a correct measurement.
The enclosing shields of the thermometer element and its attached cables should
not be electrically connected to anything other than the measuring instrument or
circuit to ensure that it is isolated. The incomplete isolation of any commercial
instrument that measures its resistance should then not cause a problem.
10.1.2 AC resistance thermometry
The connection scheme of section 5.3.5 is also suitable for lowfrequency AC
methods of measuring thermometer resistance. The capacitance of the cables shunts
the thermometer and gives it a quadrature component, so that the measuring bridge
or instrument must be capable of ignoring this and reporting only the inphase
resistance component. The dielectric loss in the cable is also a potential cause of
error, but presentday materials have a sufficiently low loss, for lengths of cables of
the order of tens of metres and for the usual measurement frequencies of 25 or
75 Hz, the effect of dielectric loss is negligible. Measuring instruments are com
monly based on the IVD ratio bridge described in section 9.1.3. Often the IVDs are
switched by relays under the control of an internal computer so that the bridge is
selfbalancing, and the computer reports the thermometer resistance directly.
10.2 Superconducting cryogenic current comparator
10.2.1 Determining the DC ratio of two resistances R
1
/R
2
This elegant instrument is the product of superconducting technology applied to the
DC comparison of two resistances. In particular, one resistance can be a quantum
Hall device, and therefore, the instrument can set up a resistance scale in terms of
the defined value of the von Klitzing constant, which represents the SI defined
ohm. Here, we only describe briefly the salient features of the instrument needed to
use it as an example of the particular problems encountered in DC measurements in
general. Figure 10.1 is a schematic diagram of a cryogenic current comparator
(CCC) circuit. More details can be found in Reference 5.
A CCC produces a nearperfect flux linkage between its ratio windings car
rying currents I
1
and I
2
and a detector winding. A fixed proportion of any residual
flux imbalance produced by the ratio windings threads the detector winding and is
transferred to the coil of the Superconducting Quantum Interference Device (SQUID)
sensor whose output, via electronics and an integrator, is applied as feedback to
adjust I
2
so as to keep the residual flux imbalance constant.
The SQUID response cycles repeatedly with changing current, and therefore, it
is necessary that the current sources track one another quite precisely during reversal
to ensure that the SQUID remains locked within one particular cycle.
Assuming for the moment that the residual flux is zero and that the voltage
difference detector D is nulled by adjusting R
A
,
R
1
R
2
N
1
N
2
N
A
R
B
= R
A
R
B
10:1
A nonzero residual flux and any offset of the voltage null detector are elimi
nated by reversing both current directions and taking the mean of the two results.
The perfection of flux linkage ensures the accuracy of (10.1) in terms of the
ratio if integer numbers of turns can be experimentally verified by a buildup
procedure in which successive larger number of turns are compared with the sum of
lesser numbers, starting with single turns. Relative accuracies of better than 1 part
in 10
9
are routine, but the stability of laboratory resistors may not be good enough
to make this meaningful.
300 Coaxial electrical circuits for interferencefree measurements
The overall values of the two current sources are governed by two independent
digitaltoanalogue convertors interfaced by optical fibre links to a controlling com
puter. DC power is supplied to them and any other electronics either by batteries
within the enclosing screen or an isolating mains power supply (see section 1.1.5).
The particular problems posed by this instrument include the following:
(i) The extreme sensitivity of the SQUID null detector to wideband interference
(up to and beyond 100 MHz). Therefore, isolation and screening must be very
carefully and perfectly carried out, and the circuit so designed as to keep the
open areas of circuit meshes as small as possible by routing goand return
current conductors together, and preferably twisting them (see section 1.1.1).
(ii) Thermoelectric effects in the room temperature circuitry (see section 1.1.11).
Reasonably rapid reversal of the current sources allows for the time constant
of these thermoelectric effects; rebalancing and taking the mean of the two
results to eliminate them is desirable.
(iii) Unfortunately, the desirability of rapid reversal is contrary to the requirement
for a settling period to allow any dielectric storage in any insulation, which
experiences a change of voltage across it to die away. Storage of charge in a
D
SQUID flux sensing, output integration
and optical isolation
R
1
R
B
N
2
N
1
N
A
R
A
R
2
CCC and associated electronics
twisted
pairs
P
H
P
H
C
H
C
H
C
L
C
L
P
L
S
L1
S
L2
S
H1
S
H2
detector output
damping
flux
linkage
Figure 10.1 Schematic diagram of a CCC for comparing the DC values of two
resistors
Application of interferencefree circuitry to other measurements 301
dielectric acts somewhat as a current source of high internal impedance, and
its effects are therefore removed with a longer time constant than that of the
innertoouter cable capacitance combined with a shunting resistance. The
current from the decay of stored charge discharges through the resistors but
avoids the comparator windings and therefore causes an error. To alleviate
this problem, the insulation material should be one that exhibits minimal
dielectric storage. PTFE is suitable. Also, the bridge sources are traditionally
current sources as shown, but there seems to be no reason why voltage
sources of negligible internal impedance should not be employed instead.
This would create an additional lowimpedance discharge path through the
comparator windings, which might be undesirable however. As a further
measure, the screens S
H1
and S
H2
in Figure 10.1 can be driven with a voltage
source, which tracks the voltage of the conductors within them. A further
advantage of the screens is that they intercept leakage currents between Hi
and Lo and divert them harmlessly back to a source, and this greatly
reduces the insulation resistance required for accuracy.
(iv) The cables to each resistor should be tightly twisted whether they are
screened twisted pairs as shown or individually screened (see section 1.1.1).
A quantum Hall device measured with this instrument can have a single connection
made either simply to each of its two potential terminals on either side of the Hall
current or to the junction point of a multipleseries connection scheme. The junction
point can either be adjacent to the device or remote from it, near to the CCC. An
advantage of the first connection is that the measurement can be made using various
pairs of potential terminals, which are not directly opposite to each other, as a check
to see whether the device satisfies the important criterion of zero longitudinal resis
tance, as prescribed in the DC guidelines. An advantage of a multipleseries con
nection is that it also greatly reduces the value required for the insulation resistance
between the Hi and Lo conductors needed to ensure an accurate result.
10.3 Josephson voltage sources and accurate voltage
measurement
Accurate comparison of two nominally equal voltage standards only requires that
they be connected back to back and that the difference voltage be measured with a
calibrated voltmeter (Figure 10.2). Because the output impedance of voltage
U
1
U
2
D
Figure 10.2 Comparing voltage sources, for example, Zener standards
302 Coaxial electrical circuits for interferencefree measurements
sources, even including the connection conductors in series with their output, is
very small, typically a few milliohms, the requirements for insulation resistance are
not demanding. To eliminate interference, the power to their components and their
temperature control needs careful isolation and should be provided by a totally
isolated power supply (see section 1.1.5). The conductors to and from the standards
should encompass as small an area as possible. Ideally, they should be a twisted
conductor pair or coaxial and be surrounded by a screen, as in the diagram. If this is
done, the detector need not be isolated. If one of the sources is a Josephson voltage
standard, the same considerations apply.
Thermoelectric emfs are a problem, albeit less so now that 10V Zener diode
voltage standards and 10V Josephson arrays are usual. To eliminate thermoelectric
emfs, the measurement can be repeated with the polarity of the sources reversed.
This can be done properly for a Josephson array by reversing its bias current, but
for a Zener standard, the best way would be if the standard had an internal reversing
relay located within the same temperaturecontrolled enclosure as the Zener itself.
Since these standards are not usually constructed this way, it is necessary to regard
their voltage as being defined by that found between two unstrained pure copper
conductors lightly clamped under their output terminals. Then an external means of
reversing the polarity needs to be provided.
One successful way of making reversing relay having very low thermoelectric
emfs is to take a relay having massive silver contacts (because it is intended to
switch heavy currents) and drill holes into the side of the contacts, which are of the
diameter of a fine pure copper wire. After the wire is inserted into the hole, a sharp
blow with a centre punch applied clear of the contact area can swage the contact
and copper wire together to make a sound pressure joint. The wires can be routed
from the contact to low thermoelectric emf terminals, or better, terminals can be
avoided altogether if the wires are left long enough to go individually into insu
lating sleeves directly to their final destination. The thermoelectric voltages in a
relay of this kind can die away in as little as 1 s to less than 1 nV following a
reversal. From time to time, the contacts need cleaning by gentle rubbing with
alcoholsoaked paper, so they would be even better if they were gold plated.
Lee et al. [6] have demonstrated an example of comparing two resistances
supplied from two independent programmable Josephson sources generating an AC
waveform at the same frequency (Figure 10.3).
Because the Josephson sources are each isolated, the resistances can be
compared as defined as twoterminalpair components with equal goandreturn
currents without the need for current equalisers. Isolation is accomplished by
powering each source with an isolated supply (see section 1.1.5) and sending the
programming instructions over an optical fibre data link. Sources, resistors and the
detector are connected by coaxial cables. The measured ratio of the two nominally
equal 10kW resistances agreed with the result from a conventional AC coaxial
bridge based on a voltage ratio transformer (see section 9.1.5) over the 0.25 kHz
frequency range of the latter, but the advantage of the programmable Josephson
junction technique is that its frequency range extends from DC to 10 kHz at
present.
Application of interferencefree circuitry to other measurements 303
There should be little difficulty in extending the technique to other values and
combinations of resistance, capacitance or inductance with another since it is
possible to generate a 90
or 180
Z
2
Z
3
Z
1
Because of the rotational symmetry of the diagrams, the transformations for the
other impedances are obtained by cyclic permutation of the subscripts.
Appendix 2
Theorem. In a network having m meshes, l links (cables) and n nodes, m l n 1.
Proof. Suppose the theorem is true for some definite value of l, m and n.
There are just five ways in which another link can be added to the network. An
unattached end of an added link is a new node.
1. One end of a new link is attached to an existing node of the network.
m !m; l !l 1; n !n 1
2. One end of a new link is attached to a newly created node in the network. The
new node divides the link it is attached to into two links.
m !m; l !l 2; n !n 2
3. A new link is attached between two existing nodes.
m !m 1; l !l 1; n !n
4. A new link is attached between an existing node of the network and a newly
created one.
m !m 1; l !l 2; n !n 1
5. A link is attached between two newly created nodes in the network.
m !m 1; l !l 2; n !n 1
In each case, the theorem is still true. But the theorem is true for a network of just
one node, m 0, l 0, n 1. Hence, the theorem is universally true.
Appendix 3
Theorem. There is exactly one cable without an equaliser joining any two nodes of
a network, that is, all the nodes of a network are joined by a simply connected
linkage of cables without equalisers. Applying the theorem of Appendix 2 to this
linkage of l
0
cables, since it has m 0, l
0
n 1.
Proof. If there were to be more than one such cable, a mesh without an
equaliser would be created.
Appendix 4
From section 6.5.12, the parameters a and b are as follows:
a
I
pk
2pbc
_ _
bK
1
hb cK
1
hc
I
1
hbK
1
hc I
1
hcK
1
hb
I
pk
2pbc
_ _
p q js t
u
_ _
A4:1
where h jwm
0
s
p
1 j=d and the Kelvin functions
K
1
hr K
1
rz j
_
_ _
jKer
1
rz jKei
1
rz A4:2
and
I
1
hr I
1
rz j
_
_ _
jBer
1
rz jBei
1
rz A4:3
such that
p p
0
bKei
1
bz cKei
1
cz A4:4
p
0
Ber
1
bzKer
1
cz Bei
1
bzKei
1
cz
Ber
1
czKer
1
bz Bei
1
czKei
1
bz
_ _
A4:5
q q
0
bKer
1
bz cKer
1
cz A4:6
q
0
Ber
1
bzKei
1
cz Bei
1
bzKer
1
cz
Ber
1
czKei
1
bz Bei
1
czKer
1
bz
_ _
A4:7
s s
0
bKer
1
bz cKer
1
cz A4:8
s
0
Ber
1
bzKer
1
cz Bei
1
bzKei
1
cz
Ber
1
czKer
1
bz Bei
1
czKei
1
bz
_ _
A4:9
t t
0
bKei
1
bz cKei
1
cz A4:10
t
0
Ber
1
bzKei
1
cz Bei
1
bzKer
1
cz
Ber
1
czKei
1
bz Bei
1
czKer
1
bz
_ _
A4:11
u
Ber
1
bzKer
1
cz Bei
1
bzKei
1
cz Ber
1
Much more than documents.
Discover everything Scribd has to offer, including books and audiobooks from major publishers.
Cancel anytime.